874 E-Bike Websites Sell to Americans: Who Runs Each

The October 2026 census of e-bike websites selling to Americans: 874 sellers read, median domain 10 years, each checked for the company it names, its registry record and the age of its domain. Verified October 15, 2026.

By Milad Ghobadibeygvand, BScN (Western University, 2014), writer and reviewer, Redtail eBikes. Verified October 15, 2026.

Two names on one checkout

On January 30, 2026, a federal court approved the sale of Rad Power Bikes' assets, yet the company's website prints one company in the contract and another in the footer. Pay that checkout and your agreement is with the name the terms print, which decides who answers a warranty claim. This page reads every e-bike website selling to Americans the same way, so you can hold any of them to the same test.

Open the terms of service on radpowerbikes.com and the legal agreement is between you and Rad Life Mobility Inc., a Nevada corporation, with notices going to an address in Deerfield Beach, Florida. Scroll to the bottom of the same page and the footer reads "© 2026 Rad Power Bikes Inc." That is the name of the company that filed for Chapter 11 on December 15, 2025 in the Eastern District of Washington.

Both lines are accurate records of different moments. A court approved the sale of the assets on January 30, 2026 after an auction in which Life Electric Vehicles Holdings Inc. bid $13,276,102 in cash, and the buyer's own letters describe what happened next. A website, a brand and a contract can each belong to a different company, and the footer is the line most likely to be left behind.

874 websites let buyers in the United States order a complete e-bike online, and the dataset lists every one. Among the 864 whose legal pages were read, 27.9% name a company anywhere on them.

The sections below read each site the way your last ten minutes before paying should: who it says is behind it, where that company is, what a government registry holds and how old the domain is.

The count: who sells e-bikes to Americans online

As of October 15, 2026, 874 websites let a buyer order a complete e-bike online and say they ship to the United States, print a US address or, for 52 Shopify stores whose pages do neither, publish a shop record that names a US shop shipping to the United States. The count comes from reading 28,216 homepages out of 57,610 candidate domains, and the dataset gives each listed site its evidence.

179 more show a price and an order control but say nothing about where they ship, or their legal pages could not be read; the dataset lists them with that status and this page does not count them. 221 more websites list e-bikes without a checkout this census could confirm, and 240 that looked like sellers sell only to other countries.

A site counts as an online seller when its own product feed or product page shows a complete e-bike, a price and a visible way to order it, and nothing on the site restricts sale to another country.

It counts toward the headline when it also gives US evidence: its legal pages say it ships to the United States, it prints a US address or, for 52 Shopify stores whose pages do neither, its public shop record names a US shop that ships to the United States. Where a store prints a US address and has a shop record, the record names a US shop for 156 of 162.

Rentals, tours, parts makers, review sites and forums are set aside, as are the 22 sites whose robots.txt asks Claude agents not to read them. The dataset lists each exclusion with its reason.

No single list holds all of these sites. The candidates came from 15 routes that do not depend on one another, among them the Common Crawl web graph, the Tranco ranking of the busiest sites, the company websites makers give Bike Index and the sites that CPSC recall notices name as points of sale. 64.9% of the sellers were found by exactly one route, so removing any route would have removed sellers that no other route found.

Of 57,610 candidate domains, 874 sell complete e-bikes online to buyers in the United States.

What kind of seller each one is

For 757 of the 874 sites the evidence settles what kind of seller each is, largest group first: 245 bike shops with a street front and an online catalog, 194 makers selling from their own stores, 143 stores of one brand that is not their maker's, 117 multi-brand retailers, 32 marketplaces or mass merchants and 26 stores of two brands. The other 117 sell online without a feed or a source that settles which kind they are.

Every label carries its basis in the dataset. If you want a brand-by-brand view, see our list of e-bike brands; if you would rather buy in person, start from our e-bike store directory.

56.1% of online sellers run on Shopify (dataset, platform column), the largest platform among them, and 95% use a .com address. Neither fact tells you anything about quality: a platform is a tool that anyone can rent.

Who is behind the checkout

Of the 864 online sellers whose legal pages were read, 27.9% name a company anywhere on them, 79.1% print a US street address and 13.5% print neither. A site that names no company has done nothing wrong by that alone; it has removed the one fact your later checks start from. Each row of the dataset quotes what its pages printed, so you can check the wording yourself.

79.1% of these sellers print a US address, and 17.8% print none at all.

Of the 710 sites that print a street address, 683 print one in the United States, 16 in Canada, 6 in China, 3 in the United Kingdom and 2 print an address without naming a country. The address is the one a site prints on its own contact, returns or terms page, read as text and kept in the dataset; it tells you where the company says it is, not where your bike was made.

What a government registry holds

237 distinct company names were searched in the open datasets of Colorado, New York, Oregon, Connecticut and Texas. Names ending in Ltd or Limited, and names beside a British or Hong Kong address, were also searched in UK Companies House (12 names) and the Hong Kong Companies Registry's open data on live local companies (23 names). The Texas file is the Comptroller's list of active franchise taxpayers.

19 sites matched a registry record by exact name and address, 55 by exact name alone and 16 by a near form of the name, and each match links its record in the dataset. A name-only match is printed as "a registry entry with this exact name exists", never as the site's company, because two companies can share a name.

Five state files are not the market. New York and Oregon publish active entities only, and the other 45 states were not searched, so a company you cannot find here may still be on file in any of them. A row with no match names the registries that were searched for its company ("none matched in CO, CT, NY, OR, TX") and never says "not registered".

The registry search for 2 names too common to scan by their words used their exact spelling under each legal form, a search that cannot see a registered name differing by punctuation inside the name.

A dealer's domain that forwards to the maker's site

42 dealer domains forward to a bicycle maker's own site (dataset, also_reached_as column), with 21 forwarding to Trek's own site and 21 forwarding to Specialized's own site.

The page you reach there prints the maker's own name and head-office address, so if you took the footer at face value you would credit a local shop with a corporation it does not belong to. The dataset lists each of those domains under the maker's row as "also reached as" and gives the dealer no operator or address of its own.

How long they have been around: three clocks

A website you are about to buy from has three ages that rarely agree: when its current domain registration began, when the Internet Archive first captured it and the year the site says it started. The median online seller's domain is 10 years old by its RDAP record, 9.6% were registered in the last 24 months and 49.9% more than ten years ago.

The first clock: the domain record

The first clock is the domain's creation date in the RDAP record that the domain registry itself keeps, read for 861 online sellers. It dates the current registration, so a domain that lapsed and was registered again shows the later date, and a business that moved to a new domain shows the new one. 36 sellers registered their domain within the last year, and the oldest records go back 35 years.

The second clock: the year a site claims

The second clock is the site's own statement. 203 online sellers print a founding year ("since 2012", "established 2009"), each quoted in the dataset. Of the 199 that print a year and have a domain record, 142 print a year earlier than their domain's current registration. That is a fact about two records, not a finding about a seller: a business can trade for years on another domain, and the registry prints both dates side by side.

The third clock: the archive

The third clock is the Internet Archive's earliest capture, which dates when a crawler first saw a working page and not when a company began.

The Internet Archive's index returned an earliest capture for 763 of the 874 sellers, holds no capture of 77 and could not be read for 34. The median earliest-capture age is 11 years.

An archive date shows a site was online by that day, and says nothing about when it began; one older than the domain's registration means the name was held earlier, and one much younger than a claimed founding year is a question for the seller.

A young domain is your reason to run the other checks, never a verdict. A seller registered last month can be a new brand with real stock, and a seller registered in 2004 can be a drop-ship front on a recycled name. The registry formation date, where a record matched, is the only one of the four dates that dates a company.

When the company behind a site changes hands or fails

Five e-bike names from the last two years show five different endings: a bankruptcy sale, a Chapter 7 filing, a withdrawal from the U.S. market, an end to U.S. sales and an asset purchase. Each is a public record you can open. None changes the rule that the warranty belongs to the company the contract names, so read the record before you rely on a brand's age.

The Rad dates do not line up, and the sources say so. The court order is January 30, 2026. The buyer's shareholder letter says the acquisition completed on February 27, 2026, and a March 5, 2026 company release announces completion. A court approval, a closing and an announcement are three different events, and a warranty question turns on the one the contract names.

One safety record sits beside the ownership record. On November 24, 2025, the CPSC told consumers to stop using three Rad battery models, citing 31 reports of fire, and printed the company's own statement that the batteries passed third-party re-testing beside it.

A notice names the company that imported the product, so search for that name as well as the brand on your bike. For how e-bike battery packs differ, see our guide to e-bike battery types.

The trap here is the footer. A "©" line, an old logo and a still-working checkout tell you the site is open; only the legal agreement tells you who you are paying.

One address, many storefronts?

32 groups of online sellers share a store, a company name, an address or an e-mail domain, and 2 more share only an analytics or advertising ID. 75 online sellers sit in one of these groups. Only the first three kinds describe a common operator, and the dataset names each group in its marker_group column; the fourth never does, because the same agency or template can put one analytics or advertising ID on many companies' sites.

The largest identity group is 12 online sellers that print the name California Electric Bikes and an address in Fair Oaks, CA: caebikes.com, d2d393-33.myshopify.com, ebikesafetyclasses.com, ebikeshoptemecula.com and others. The largest store group is 6 domains served by one Shopify store: ebikes30a.com, pedego.com, pedegodanapoint.com, pedegoelectricbikes.com and others.

The census never writes "same operator" on a shared tool. It prints the observation (both pages embed the same property ID), and you draw the conclusion.

Marketplaces and big-box stores

Of the 100 largest US retailers on the National Retail Federation's 2025 list, 19 show e-bikes for sale online, 46 showed none on the routes this census could read, 4 could not be read, 1 appears only in federal recall notices and 30 have no single consumer website to test. The list is compiled by Kantar and printed on nrf.com; the verdicts come from three routes per retailer, each dated.

19 of the 100 largest US retailers show e-bikes for sale online.

Four routes decide a row. The retailer's own site search was run for "electric bike" in a browser, the search engine's index was asked for e-bike pages of that domain, the e-bike page the index returned was fetched and read where the retailer's site allowed it, and where a retailer's product data had been read on September 6, 2026, the listing count is shown.

Of the 32 mass merchants and marketplaces on the curated list, 11 rest on a page of their own site that we read, 2 on product data read on September 6, and 19 on a search-index listing alone, because every route we tried met a bot wall; each is a row of the dataset.

A "none found" means the routes that answered returned no e-bike for sale, and the list says how many answered; it never says the retailer does not sell one. A page returned by a search index is a lead that the retailer lists e-bikes, not a read of its checkout.

Marketplaces beyond the top 100

What federal recall notices say about points of sale

The CPSC's e-bike recall notices name 35 websites as places a recalled e-bike was sold, across 25 of the 42 e-bike recalls the agency has published since 2001. Amazon is named in 3 of those notices, Walmart in 3, AliExpress in 2 and eBay in 2. A notice records where specific units were sold in a stated period, so it dates a sale point and does not describe today's shelf.

Amazon and eBay ask Claude agents not to read their sites (Amazon's robots.txt names them), so they appear here from these notices alone. If you want one recall read in full, see our explainer on the Lectric recall.

When a marketplace lists a bike, the seller of record can be a third party, and the marketplace's own legal name is not the seller's. Read the line that names the seller before you read the star count, and search the CPSC recall list for the brand before you pay. For more on where to buy, read our guide to the best places to buy an electric bike.

How to check an e-bike website in ten minutes

Ten minutes and six open tabs are enough for you to learn whether a stranger's checkout belongs to a company that exists. Start with the company the contract names, confirm it in a government registry, date the domain, read the archive, search the CPSC and pay in a way the law protects. Every step ends at an official page.

Skip them, and if the bike never arrives or the battery fails, you may not know which company to hold to the warranty.

The ten-minute checklist

• ☐ The company named in the terms of service, copied exactly
• ☐ That name in a government registry, with its status and formation date
• ☐ The domain's creation date in ICANN Lookup
• ☐ The earliest Internet Archive capture, set beside the domain date and any founding year
• ☐ A CPSC search for the brand and for the company
• ☐ A credit card for the payment, and the order page, terms and warranty saved

One: the company the contract names

Open the terms of service, the privacy notice and the contact page. Find the sentence that begins "these terms are a legal agreement between you and". The name after it is your counterparty, whatever the footer or the logo says. A site with no company name on any of those pages is not proof of anything, but it removes the one fact your later checks start from.

Two: that company in a government registry

Search the exact name where the site says it is registered. The National Association of Secretaries of State links every state's business search; Companies House covers the United Kingdom. Compare four things: the exact name, the status, the formation date and the address. A match on the name alone gives you a lead, never a finding.

Three: the domain's creation date

Paste the domain into ICANN Lookup and read the creation date and the registrar. That date belongs to the current registration, so a transfer or a lapse can make an old business look new, and a new store can sit on an old domain. Read it beside the next check, never alone.

Four: the archive

The Internet Archive shows when pages from the domain were first captured and what they said. An earliest capture older than the domain's creation date means the name was held before the current registration. An earliest capture a few weeks old on a site that claims a decade is a question for you to put to the seller in writing.

Five: the recalls and warnings

Search the CPSC's recalls and product safety warnings for the brand and for the company named in step one, because a notice may name the importer rather than the brand. A result shows a maker that corrected something in public or one that has not; read what it says before you read anything into it.

Six: pay in a way the law protects

Two federal rules do most of the work for an online order. The FTC's Mail, Internet, or Telephone Order Merchandise Rule requires a seller to ship within the time it states, or within 30 days if it states none. When it cannot, the seller must offer you a choice: accept the delay or cancel for a prompt refund.

The Consumer Financial Protection Bureau's Regulation Z treats a credit card charge for goods "not delivered to the consumer … as agreed" as a billing error, which you dispute in writing within 60 days of the statement that shows it.

Who enforces what when an order goes wrong

The table takes six ways an e-bike order can go wrong and names the body that owns the lever for each. A late bike goes to the seller and the FTC's shipping rule, a charge for a bike that never came goes to your card issuer within 60 days, and a battery that fails goes to the CPSC. Each row links the rule that creates the lever, so you can read it yourself.

A written warranty is the promise of the company named in the terms, and the table cites the FTC rule on warranty terms. What the warranty covers, who pays freight on a claim and what voids it are in the warranty itself, so save a copy the day you order.

What a record cannot tell you

A registry entry, a domain date and a footer prove that a company exists, when a name was registered and where an address is printed. They never prove that a bike will arrive, that a warranty will be honored or that a price is fair. Read every field in this census as the start of a question you put to the seller in writing.

Four readings are easy to get wrong:

• A missing company name is a fact about the page and tells you nothing about the seller.
• A shared analytics ID is a fact about two pages and does not make one owner.
• A registry match on a name alone is a lead, because two companies can share a name in different states.
• A young domain is not a young company, and an old one is not an old company.

How much of the market this census saw

A list of websites is only as good as its misses, so the census measured its own. After the candidate list was built, 44 purchase-intent queries observed in Google's autocomplete were put to a search engine and the result sites were classified.

15 of the 18 sellers among them were already in the candidate list through a route other than search, a recall of 83.3% with a 95% interval of 60.8 to 94.2%. That measures recall against the sellers a search engine shows for buying questions, not against the whole market, and the sellers it found that the routes had missed were added to the registry.

A second round of 21 queries (9 about buying online, 12 about stores in US cities) was run against the list as it then stood. 15 of the 16 sellers among the results were already in it, 93.8% with a 95% interval of 71.7 to 98.9%.

Checked against the busiest sites

Every one of the 22,301 sites in the Tranco top 30,000 that does not use a country-code domain was a candidate whatever its name, so that mass merchants without a bike in their name were not skipped. 19,840 were fetched, 12,637 homepages were read and 13 turned out to be e-bike sellers.

Checked against names the routes were not seeded with

A second, smaller probe used names the routes were not seeded with. 124 e-bike makers and retailers a US buyer might know were typed from memory, which makes the list a probe and not a sample of the market, and checked against the census.

The census already held 106 of them. Of the 18 that were not, 7 did not resolve (a wrong guess or a closed site), 5 could not be read and 5 showed e-bike content when read at their sites, of which one turned out to be an e-bike seller with US evidence and was added.

The sites the census could not classify

The census could not classify 17,042 sites: their homepage could not be read, or it was read and the site search or product feed the census needed could not be. A parked page is a page that was read, and is not counted here. The sites fall into two groups.

3,798 sites from routes that point at e-bike sites on purpose, such as the company websites makers give Bike Index and the sites that recall notices name, first answered with a wall, a blank page built by JavaScript, an error or no answer. Each got a second try, a browser render and a read of its sitemap: 1,831 were read that way, 46 turned out to be parked pages and 1,921 stay unread.

With 85 more whose homepage was read but whose search or product feed could not be, 2,006 sites from these routes stay unclassified, and the census says nothing about them.

The other 15,030 came only from the two bulk routes, the ranking of the busiest sites and the Common Crawl link graph. They are too many to read one by one, so the two largest kinds were sampled.

A random sample of 228 of the 5,719 that answered with a page built by JavaScript was rendered in a browser, and none of the 130 that rendered was an e-bike seller; at the 95% level that bounds the group at 166 sellers or fewer (an interval of 0 to 2.9%), on the assumption that the pages that would not render resemble the ones that did.

A random sample of 223 of the 1,701 that answered with a wall was rendered in a browser, and none of the 58 that rendered was an e-bike seller; at the 95% level that bounds the group at 105 sellers or fewer (an interval of 0 to 6.2%), on the assumption that the pages that would not render resemble the ones that did.

The remaining 7,610 bulk-route sites answered with an error or no answer at all, or kept their search or product feed closed. They were not sampled, so the census says nothing about them either.

6 more sites whose route was not recorded stay unclassified too.

What an AI shopping assistant can read on each site

529 of the 874 sellers publish at least one file written for AI shopping agents, read on October 16, 2026: 499 publish a Universal Commerce Protocol profile at /.well-known/ucp, 523 an llms.txt and 482 an agents.md. 36 sites could not be read for at least one file, and the companion file carries all three readings for every row of the dataset.

Half of shoppers now use an AI assistant at some point in a purchase, Salesforce reported in July 2026, and 65% of the 403 US consumers Clutch surveyed in December 2025 had used one to research a product before buying. An assistant reads what a site publishes for machines, so this census read the three files an agent looks for first on every listed site.

The three files, counted

The reads that could not tell either way, counted apart from the shares above: 23 for the profile, 24 for llms.txt and 35 for agents.md, on 36 sites in all.

The platform decides almost all of it. 482 of the 490 sellers on Shopify publish the profile, because Shopify generates an agents.md file "by default for every store" and serves its text at /llms.txt unless the store overrides it; every one of the 482 agents.md files read here carries that generated text. Among the 384 sellers on every other platform, 17 publish a profile and 39 an llms.txt, most of those written by a plugin.

What the files say, and what they leave out

A generated agents.md names the store as the shop calls itself, lists its policies and tells the agent where checkout is. It does not name the company behind the checkout: of the 164 sellers whose own pages name a company and whose agents.md was read, 13 name that company in the file. No file read here carries a registry number, a domain creation date or an archive capture.

Those are the records in the dataset, which is why it is published as a CSV and, from October 16, 2026, as JSON that an assistant can read in one call.

The assistant's answer about a seller rests on the pages it could read, and buyers are following those answers: Adobe Digital Insights measured AI-referred visitors converting 42% better than other traffic on US retail sites in March 2026, with 39% of the more than 5,000 US consumers it surveyed that month saying they had used an AI assistant for online shopping.

How to use this census with an assistant

Two-thirds of US shoppers (67%) told Commerce and PayPal's researchers in spring 2026 that they were interested in trying an agentic shopping tool, and across the three countries surveyed they wanted it for research and comparison, with purchase confirmation, fraud protection and payment security the features that would make them use one. The same division of labor works here.

1. Hand it the dataset. Give the assistant the CSV or the JSON and the site's domain, and ask for the company the terms name, the registry match and the three dates.
2. Make it quote the terms. Ask it to open the site's terms of service and quote the company name; a name it cannot quote is a name the site does not print.
3. Keep the payment. Pay by credit card yourself for the billing-error right in step six, and confirm the order yourself.
4. Treat "legit" as a lead. Run the six checks on whatever it reports, and search the CPSC recall list for the brand.

Every e-bike website, one row each

The registry lists 874 websites that sell e-bikes online and say they ship to the United States, print a US address or, for 52 Shopify stores whose pages do neither, publish a shop record that names a US shop shipping to the United States, grouped by the kind of seller each is, and every row is in the dataset.

Each row shows you the site's own name, the company it names and where, what a government registry holds for that name and the age records.

A dash or an empty field means the source did not print it or the census could not read it, never that the thing does not exist, and "none matched in CO, CT, NY, OR, TX" lists the registries that were searched (dataset, registries_searched column) and never says the company is unregistered.

The same rows, with every source, are in one CSV, which also holds the 179 online sites that show prices and an order control but state no US shipping and print no US address. Each row records the last day the site was read and the page that showed its order control, and 16 rows carry a flag because every product title the census counted for them reads as an electric motorcycle, dirt bike, moped or scooter.

Redtail sells e-bikes from 16 of the sites listed here. Those rows carry identity fields only (what the site prints about itself and any registry match by name and address) and no age records, so that this page does not grade its own suppliers. 19 rows have withheld fields, each named in the dataset.

Bike shops with a website (245)

Makers selling from their own store (194)

Single-brand stores (143)

Multi-brand retailers (117)

Online sellers whose kind is not determined (117)

Marketplaces and mass merchants (32)

Two-brand stores (26)

Our own row

Redtail eBikes sells e-bikes at redtailebikes.com, so this census profiles that site by the same method as every other. Its row names the company that operates it by reference to our terms of service, where the contract is printed, and carries the same age and platform fields. A buyer's guide that reads every seller should publish its own seller's record beside the others, so you can check it.

Buying your e-bike from us

If you run the six checks on us, this is what you will find, and every line of it is printed on our own pages. The company that answers for your order is named in our terms of service, and a person answers for it too: the phone line on our contact page, where every voicemail is returned, and support@redtailebikes.com, answered within one business day.

Every e-bike we stock ships free within the contiguous United States, and our shipping policy states the clock that step six asks every seller to state: dispatch within 1 to 3 business days of confirmation, and delivery for most e-bikes within 5 to 10 business days after dispatch. If a box arrives damaged, note it with the carrier and email us photos within 48 hours of delivery so we can resolve it, as our refund policy says.

Checkout takes Shop Pay, Apple Pay and Google Pay as well as the major cards, so you can pay by credit card and keep the billing-error right described in step six. Your bike is covered by its maker's warranty where applicable, and we coordinate the repair or replacement remotely, as the refund policy says.

If you shop with an AI assistant, our store publishes the three files the assistant section counts, so it can read our prices, stock and policies through the Universal Commerce Protocol profile at redtailebikes.com/.well-known/ucp and prepare a checkout for you. Our row sits in the dataset beside every other seller's, with the same fields.

If the e-bike you want is one we do not stock, run the same ten-minute check on the site that sells it before you pay. The checks work the same on any checkout, ours included.

Frequently asked questions

Which e-bike websites are legit?

This census lists 874 websites that sell e-bikes to US buyers and rates none of them. A rating would claim more than a public record can show. What the record does show is who each site says is behind it, what a government registry holds for that name and how old the domain is, and the registry and the dataset print those for every row. Run the six checks on any site before you pay.

Which e-bike brands are the most trusted?

This census does not rank brands or sites: it records who each site says is behind it, what a registry holds for that name and how old the domain is. For a brand-by-brand comparison, see our guide to the best e-bike brands, and for single models, our e-bike reviews.

How can I tell if an e-bike website is legit before I buy?

Find the company the terms of service name, search that exact name in a government registry, date the domain, read the archive's earliest capture, search the CPSC for the brand and the company, and pay by credit card. 13.5% of the online sellers whose legal pages were read print neither a company name nor a street address (dataset), which leaves a buyer the registry and the domain date. The ten-minute check links every step.

How do I buy an e-bike online safely?

Check the seller before the bike. Buy from a site whose terms of service name a company, pay by credit card so a bike that never comes is a billing error you can dispute, and save the order page, the terms and the warranty the day you order.

If the seller states no shipping time, the FTC's mail-order rule gives it 30 days, after which it must offer you the choice to wait or cancel for a prompt refund. Our first-time e-bike buyer questions cover choosing the bike itself.

Is an e-bike website a scam if it names no company?

Not by that alone. 13.5% of the online sellers whose legal pages were read print neither a company name nor a street address, and every one of them is listed here because it takes orders (dataset). What a missing name takes away is the fact every other check starts from, so ask the seller in writing which company you are paying before you pay.

What should I do if an e-bike website looks fake?

Report it to the FTC at ReportFraud and to your state attorney general, whom Find my AG lists. If you paid by credit card and the bike never came, dispute the charge in writing within 60 days of the statement that shows it (12 CFR 1026.13). The levers table names the body for each kind of problem.

Can an AI shopping assistant check an e-bike website for me?

Partly. An assistant reads what a site publishes for machines, and 529 of the 874 sellers in this census publish at least one of the three files written for agents; a generated file names the store, not the company behind the checkout. Hand the assistant the dataset for the registry record and the domain and archive dates, keep the payment and the final confirmation yourself, and run the six checks on whatever it reports.

How many e-bike companies and websites are there in the United States?

874 websites let buyers in the United States order a complete e-bike online and show US evidence on their own pages or in their shop record, and the dataset lists each one. 179 more show a price and an order control without saying where they ship, and 221 list e-bikes without a checkout this census could confirm.

Those sites name 237 distinct companies, and a company can run several sites, so websites outnumber companies and a count of sites is a floor rather than a census of firms.

Does Walmart sell e-bikes?

Yes. Walmart's own search for "electric bike" returned adult e-bike listings in a browser read on October 14, 2026, and a read of its product data on September 6, 2026 counted 72 e-bike listings.

Best Buy and Costco listed e-bikes in product data read on September 6, 2026, and a search engine's index lists e-bike pages of Target, The Home Depot and Lowe's that their own sites would not show this census. The table of the 100 largest retailers shows which others do and which could not be read.

Is it safe to buy an e-bike on Amazon, AliExpress or Alibaba?

The record this census can read is the CPSC's: 25 of the 42 e-bike recall notices it has published since 2001 name a website where the recalled bike was sold. Amazon is named in 3 of those notices, Walmart in 3, AliExpress in 2 and eBay in 2.

So a listing on a marketplace can be a recalled product. Search the CPSC recall list for the brand before you order, pay by credit card for the billing-error right and read which company is the seller of record.

Who owns an e-bike brand, and how do I find out?

The company that owns the contract is the one the terms of service name, and the brand, the domain and the footer can each belong to someone else, as Rad Power Bikes' terms and footer show. 27.9% of the online sellers whose legal pages were read name a company anywhere on them. Search the exact name in the state where it says it is registered, then compare the status, the formation date and the address.

Where is an e-bike company located, and is it registered?

Where its own legal pages say it is, and a government registry is how you confirm it. 683 of the online sellers that print a street address print one in the United States, 16 in Canada and 6 in China, and a registry record matched 19 sites by name and address (dataset). Search the exact company name in the state the site names through the NASS directory of state business searches.

Are all e-bikes sold online made in China?

The records this census read cannot say where a bike is made. They say where a company says it is, and 6 of the online sellers that print a street address print one in China (dataset); a factory appears on a seller's page only when the seller chooses to print it. If it matters to you, ask the seller in writing where the bike was built and keep the answer.

How long has an e-bike company been in business?

A domain's creation date, the earliest archive capture and the year a site claims are three different clocks. The median online seller's domain is 10 years old and 9.6% were registered in the last two years. Of the 199 sellers that claim a founding year and have a domain record, 142 claim a year earlier than the current registration. A registry's formation date dates the company, and the age section explains the rest.

What happens to my warranty if an e-bike company goes out of business?

A written warranty is a promise of the company the terms name, and a bankruptcy sale can move the brand and the website to a different company that decides whether to honor it. Rad Power Bikes is the public example: the court approved the sale of its assets on January 30, 2026, and the current terms name a different company than the footer does. Read the record and save the warranty text on the day you order.

How do I know if an e-bike company has gone out of business?

Look for a court record, not a rumor. A bankruptcy petition or a sale order is public on the federal court's docket, and a sold brand's terms of service can name the new owner, as Rad Power Bikes' sale order and current terms show. The CPSC's recall and warning list names the importer, which can be a different company than the brand, and five endings with their records are on this page.

Are cheap e-bike websites selling the same bike under different brand names?

32 groups of online sellers share a store, a company name, an address or an e-mail domain. The dataset names each group. The public record cannot see a factory, so a design shared behind different names is outside what a registry or a page shows. The clusters section explains which overlaps count as one operator and which, like a shared analytics ID, never do.

The bottom line

You can check any e-bike website in ten minutes, and what you find decides more than the price does. 874 sites sell complete e-bikes to Americans online, 27.9% of those whose legal pages were read name a company, and the domain of a typical seller is 10 years old by its RDAP record. Use the six checks, then pay by credit card and keep the warranty text.

Key takeaways

• Read the terms of service for the company name before the footer, because the contract names who answers a warranty claim.
• Search that exact name in a government registry, and treat a name-only match as a lead.
• Date the domain, compare it with the archive and with any founding year the site claims, and ask the seller in writing about any gap.
• Pay by credit card so the billing-error right in 12 CFR 1026.13 applies.
• If an assistant checks a site for you, hand it the dataset: 529 of 874 sellers publish a file written for agents, and none of those files carries a registry record or a domain date.

Check your state's e-bike rules before you order, compare brands in our guide to the best e-bike brands, and shop commuter e-bikes or every e-bike we stock once a seller clears your check.

Sources, limitations and how this page was verified

Every number on this page is computed from saved files by a script and re-derived by a second script that shares no code with the first, and the dataset holds the rows they count; the figures in the verified ledger are the ones that drive a decision.

A court order, a CPSC notice, a company's own terms page and a domain registry's record each have a single publisher, so the reference list names no second source for them. Sites were read on October 14 and 15, 2026, and every row of the dataset carries the last day a saved record of that site was written.

The verified ledger

Limitations

1. Scope of "sells to Americans." A site is counted when its own pages show a complete e-bike, a price and a visible order control, nothing restricts sale to another country, and it gives US evidence: its legal pages say it ships to the United States, it prints a US address or, for 52 Shopify stores whose pages do neither, its public shop record names a US shop that lists the United States among its shipping destinations. A site that ships to the US without saying so cannot be told apart from one that does not; 179 such sites (every one of them a site whose legal pages were read and say nothing either way) are in the dataset, flagged, and not counted. A store outside the United States that lists the United States among its destinations is not counted on that alone: a destination list says where a shop will send an order, not that it sells to Americans.
2. Where and when the sites were read. Sites were read on October 14 and 15, 2026, from a connection in Canada. Many Shopify stores quote Canadian dollars to a Canadian visitor and some brand sites send one to a Canadian storefront, so a price label on a page read from here can be Canadian dollars on a store whose US shoppers see US dollars; the census takes a shop's currency from the shop's own record and prints a price range only where that record names US dollars. 25 brand sites that served a Canadian storefront to that connection were read again in their US form; where that failed, 22 online sellers' legal pages are the Canadian storefront's, and their rows say so. A store that treats a Canadian visitor differently in some other way could be under-read. A price in euros, pounds or rupees counts against a site only where its legal pages were read and none says it ships to the United States.
3. Registry coverage. Every company name was searched in five state datasets, and 12 names in UK Companies House and 23 in the Hong Kong register. New York and Oregon publish active entities only, Hong Kong's open data covers live local companies only, and 45 states were not searched, so "none matched" never means "not registered." Names too common to scan by their words (2 of them) were searched by exact spelling only.
4. Domain age is the age of a registration. The RDAP creation date dates the current registration, and a transferred, lapsed or re-registered domain can show a date that is younger or older than the business. The Internet Archive's index returned an earliest capture for 763 of the 874 sellers, holds no capture of 77 and could not be read for 34.
5. Unread sites. 16,667 homepages could not be read: walls, timeouts, dead hosts and pages built by JavaScript. The recall section sizes what a random sample of them held.
6. Search-index evidence is a lead. For 19 of the 32 mass merchants and marketplaces on the curated list, every route we tried to read the retailer's own pages (three sets of request headers, a real browser, the site's own search box) met a bot wall or an error, and a page returned by a search index shows that the retailer lists e-bikes, and whether its checkout works today stays unknown.
7. Kinds are inferences. A site's kind comes from its feed vendors and the route that found it; 117 sellers have no feed or source that settles it.
8. Names and addresses are as the sites print them. No natural person is named. A street address is printed only where the operator carries a company form and a registry matched its name and address; otherwise only a city and state or country appears.
9. Name-only matches are leads. Two companies can share a name in different states; a registry entry with the exact name is printed as such and never attached to a site by name alone.
10. Recall has a bias. The audit measures the share of search-visible sellers that the routes had already found; sellers that rank poorly for buying questions are under-represented in it.
11. Brands Redtail sells. Their rows carry identity fields only, which is why their order_evidence column is blank.
12. A snapshot is not a rating. Nothing here ranks, grades or recommends a site.
13. How complete e-bikes are counted. A product feed's complete e-bikes are recognized from each product's title (and, on Shopify, its product type) by two written rules that must both accept it, and "e-bikes listed" counts the products they accept at $250 or more. A product that only the looser rule accepts, such as a "fat tire" bike whose title also lists a motor and a battery, is a lead and not a count, and the store is confirmed by reading its product pages instead. Shopify catalogs are read to 2,000 products and WooCommerce catalogs to 1,000, so a longer catalog is counted over the pages read. It is a floor, and counts are not comparable between sites. A WooCommerce store that answers its product list with a bot challenge was read through a real browser, to 300 products; stores that could not be read either way (1 of the listed sites) keep the count of an earlier product search made with one rule. A store that lists its bikes under model names alone is recognized instead from the priced products in its feed or its own e-bike page, by reading up to three product pages for an e-bike title or electric category, a price and an order control, and shows no count.
14. Sites whose e-bike page could not be read. 80 of the 3,220 sites that mention e-bikes without a visible purchase path could not be read on their own e-bike page after two tries. They are counted as unread, never as non-sellers.
15. Sites whose homepage never says e-bike. 15,271 homepages read fine, mentioned no e-bike and offered a search that a plain fetch could not run. The 2,647 that carry a bike word in their domain, name, title or description were all read through their sitemaps (645 sitemaps read, 0 sellers among them), and a random sample of 400 of the rest was read the same way (166 sitemaps read, 0 sellers). A shop that lists one or two e-bikes and publishes no sitemap would not be seen this way.
16. What counts as an order control. A site is online when a product page reads as a complete e-bike with a price and a visible order control. The control is read from the page's markup, not from its scripts: a visible button, link or input whose text begins with an order phrase such as "Add to cart", "Buy now", "Pre-order" or "Reserve now", beside a price, counts. A financing banner, a terms page, a menu tile, a product title, a promo line, a link to a marketplace and a form hidden from view do not, and a dealer locator makes a store catalog-only. A page that no longer leads to the product, such as a home page or a collection, is no reading. Of the 394 listed sites whose verdict rests on a product page, 382 show such a control on a page we parsed (20 only after a real browser had drawn the page), 5 show their own buy button saying sold out or notify me, 6 stand on a page a person read in a browser and 1 on an earlier reading because its pages could not be read again. Where a store's product pages showed a dealer link and no order control, the store moved to catalog-only (4 sites); where they showed nothing at all, it moved to unverified (39), and a page whose control appears only after a size or option is chosen reads that way too.
17. What counts as an e-bike page. The page's own title, or a model name on the site's e-bike category beside the word electric on the page itself or in the page's own category path (its breadcrumb), decides what a complete e-bike is. A product page that says in-store only, pick-up only or not available for direct shipping is set aside, and an invitation to call or visit beside an order control is not; a part, a drive kit, a subscription, a corporate page, a blog post or a vehicle that is not a bicycle (a mini jeep, a ride-on toy) is never an e-bike page. Where a product tile's link says only "Shop now", the product is named from the link's address. A category or brand index is never taken for a product page: a page whose address says category, brand, collection, archive or filter, or whose name is made only of category words, is read for the product tiles it holds (or the site's sitemap is searched for e-bike product addresses), and a site whose only evidence is such an index is counted as listed and unverified. A store that lists its bikes under names with no link to a product page, or builds its pages with JavaScript and publishes no product feed, is counted as listed and unverified, never as a non-seller.
18. The second look and the feed look. Where the first look found no control, the census read up to three further e-bike pages of the same store (the in-stock products of its own product feed, then the product links on the pages already read, such as the variant pages behind a model page that shows only a store locator, then the electric category listing of a store whose pages print a cart form) and counted a control only beside a price on that page and not beside the page's own sold-out wording: 145 stores were given that second look, 37 of them showed a visible control on a further page, and 10 of the page-route sellers listed here stand on it. Stores counted through their product feed (a Shopify or WooCommerce product list shows that a product exists, has a price and an in-stock flag, not that the shop takes an order) were checked the same way: up to two complete e-bike pages of each were opened in a browser, in-stock products first. 717 such stores were looked at; 568 showed a visible order control beside a price, 28 showed their own sold-out button, 36 US-evidenced stores whose pages showed neither (a page that says call or email us, or in-store purchase only) moved to unverified, 6 to catalog-only, and for 64 no usable page could be read, so the feed's class stands for them. Stores that draw their product tiles and order controls with script, so that a plain request returns an empty shell, were read a second time in a real browser; a maker site that picks its storefront by the visitor's country from the first page a session asks for (trekbikes.com does) was asked for its US path first, so that the product pages cited carry US prices; 258 stores were read that way and 28 of them now stand among the listed sellers.
19. A cart form is not an order control. One platform used by independent bike shops prints an "Add To Cart" form on every product page, whether or not the shop takes that order online; its pages draw a visible "Add to Cart" button only for a bike that can be ordered. 48 sites showed only the form on the pages first read. A second look at further e-bike pages, taken from the electric category listing of each store, found a visible order control for 10 of them, which are listed; of the rest, 32 show US evidence and stay on the list as unverified, because a store that sells a bike only in person can show the same page.
20. Controls the checks do not recognise. The check reads a button, link or input whose text begins with an order phrase, and a control drawn another way can be missed. One of the 60 sites in the unverified sample showed a button whose label is drawn twice, a visible "Buy" and a hidden copy for an animation, which the check read as "BuyBuy". The check now ignores the hidden copy, and the 160 stores whose pages had shown no control were read again with it: none showed a control on an e-bike page; one showed a control on a bike that is not electric, picked from the site's menu, which does not count. A control that appears only after a size or option is chosen, or only to a logged-in visitor, can still be missed, and the store reads as unverified.
21. Model pages that never say electric. A bike shop whose e-bike pages carry a model name and nothing else, with no electric word in the title, description or heading and no category path (a dealer's page for a model known only by its name), reads here as not selling e-bikes online. The count is a floor for that reason.
22. Robots files. The census read every listed site's robots.txt and left out the 22 sites whose file asks Claude agents to stay out, together with any domain that forwards to one. The recall-notice table can still name such a site where a CPSC notice lists it as a point of sale: the notice is the source, and nothing was read from the site. No route could read the robots.txt of 2 listed sites and of 6 of the 69 top-100 retailers that map to one website, because a bot manager answers even a real browser with a challenge; a listed row is published from what a visitor sees and its robots_txt field says COULD-NOT-LOOK.
23. An order control is not a delivery promise. The census checked that an e-bike page shows a price and a way to order, not that every model ships to every address. Some counted sites mark a model for pick-up or special order, and the dataset's order_evidence column records the control and the page it was seen on.
24. Precision. A random sample of 150 sites was drawn from the 901 then listed, and a person opened each one's evidence page. Of the 150, 4 have since come off the list, so 146 are scored; they are a random sample of the 868 listed sites that were also on the list the sample was drawn from. Of the 146, 136 showed a complete e-bike offered for sale online, none did not and 10 could not be settled, a precision of 93.2% with a 95% interval of 87.9 to 96.2%, counting every unsettled row against it. The other 6 listed sites were added after the sample was drawn, and a person opened each of them before adding it. The sample asks whether the evidence supports the label, not whether a checkout completes, and it audits a list whose evidence a person had already read in full. Three samples have been drawn in all. The first found a category page standing as a product page, the second a product-feed row whose pages show no order control and the third a cart form that one platform prints on every product page; each rule was repaired and swept across every saved page, and the sites the third repair took off the list are not scored. A reader who finds a row that does not belong should write to the correction address below.
25. The unverified group. The list can also err by leaving sites out. Another 221 websites list e-bikes without a checkout this census could confirm, 146 of them with US evidence. A random sample of 60 of the 106 that were unverified and showed US evidence when the sample was drawn was opened by a person: 6 sold complete e-bikes online, 35 did not and 19 could not be settled. Those 6 are now on the list. The sample puts the share at 10.0% (95% interval 4.7 to 20.1%, counting every unsettled site as not selling), which suggests about 5 more (2 to 9) among the 46 sites the sample did not draw, and none of those has been opened.
26. Electric motorcycles under an e-bike label. A seller's own product titles decide what counts as a complete e-bike, and some sellers file electric dirt bikes and motorcycles under that label. A sweep of the counted titles of 447 sites counted through a product feed and 387 counted through a product page found 16 (5 and 11) whose titles all read as electric motorcycles, dirt bikes, mopeds or scooters. They stay listed because the seller sells them as e-bikes, and the dataset marks them in its moto_titles_only column.
27. Agent files were read once, on one day, from one connection. On October 16, 2026 each site's /.well-known/ucp, /llms.txt and /agents.md were requested once with a plain user agent, and once more with a browser user agent on HTTP/1.1 when the first read met a wall or got no answer. 36 of 874 sellers could not be read for at least one file and are counted apart, a file served only to a named agent reads as absent here, and a platform can add or remove these files for every store it hosts in one release.
28. "Names the company" in an agent file is a text match. The company a site's pages name was searched in its agents.md and llms.txt as printed, so a file that names the company in another form (an abbreviation, a trade name) counts as not naming it.

When this page expires

This page is a snapshot of the sites as they were read on October 14 and 15, 2026. A site can open, close or change owners at any time, and a registry or a domain record can change the day after it was read. Re-check any row before relying on it, and treat a figure older than a quarter as a prompt to run the six checks again.

The agent-file counts were read on October 16, 2026, and a platform can switch those files on or off for every store it hosts in one release, so re-read a site's files before quoting a share.

Corrections go to support@redtailebikes.com, and a confirmed error is corrected in the body with its date.

Sources

T1 · Government records and primary legal text

1. Federal Trade Commission (codified by the Office of the Federal Register). Mail, Internet, or Telephone Order Merchandise, 16 CFR Part 435. § 435.2(a)(1), (b)(1); current text as of October 9, 2026; read October 14, 2026.
2. Consumer Financial Protection Bureau (codified by the Office of the Federal Register). Truth in Lending (Regulation Z), 12 CFR § 1026.13 Billing error resolution. § 1026.13(a)(3), (b)(1); current text as of October 9, 2026; read October 14, 2026.
3. Federal Trade Commission (codified by the Office of the Federal Register). Rules Governing Disclosure of Written Consumer Product Warranty Terms and Conditions, 16 CFR § 702.3. § 702.3(c); current text as of October 9, 2026; read October 14, 2026.
4. U.S. Consumer Product Safety Commission. Recalls (the agency's recall list, read through its public API). 42 e-bike recalls among 10,047 records; first September 18, 2001, last October 1, 2026; issued October 1, 2026 (latest record); read October 14, 2026.
5. U.S. Consumer Product Safety Commission. CPSC Warns Consumers to Immediately Stop Using Batteries for E-Bikes from Rad Power Bikes Due to Fire Hazard, Risk of Serious Injury or Death. warning of November 24, 2025; issued November 24, 2025; read October 14, 2026.
6. U.S. Bankruptcy Court, Eastern District of Washington. Rad Power Bikes Inc., voluntary petition for relief under Chapter 11, Case 25-02183. Doc. 1; issued December 15, 2025; read October 14, 2026.
7. U.S. Bankruptcy Court, Eastern District of Washington. Notice of auction results, In re Rad Power Bikes Inc. and Summit Collective Inc.. Doc. 144; issued January 23, 2026; read October 14, 2026.
8. U.S. Bankruptcy Court, Eastern District of Washington. Order approving the sale of assets, In re Rad Power Bikes Inc. and Summit Collective Inc.. Doc. 180; issued January 30, 2026; read October 14, 2026.
9. U.S. Bankruptcy Court, Central District of California. Electric Bike Company, LLC, voluntary petition under Chapter 7 (Case 8:25-bk-12810-SC) and Schedule A/B. Doc. 1 and Doc. 1-5 (Schedule A/B, Part 10 line 61); issued October 3, 2025; read October 14, 2026.
10. ICANN. ICANN Lookup (RDAP domain registration data). creation event, registrar; read October 14, 2026.
11. Colorado Department of State (Colorado Information Marketplace). Business Entities in Colorado. dataset 4ykn-tg5h, all statuses; read October 14, 2026.
12. New York State Department of State (Open NY). Active Corporations: Beginning 1800. dataset n9v6-gdp6, active entities only; read October 14, 2026.
13. Oregon Secretary of State (Open Data). Active Businesses - ALL. dataset tckn-sxa6, active businesses only; read October 14, 2026.
14. Connecticut Secretary of the State (CT Open Data). Connecticut Business Registry - Business Master. dataset n7gp-d28j, all statuses; read October 14, 2026.
15. Texas Comptroller of Public Accounts (Texas Open Data Portal). Active Franchise Taxpayers. dataset 9cir-efmm (taxpayer name and Secretary of State file number); read October 14, 2026.
16. Companies House (UK). Find and update company information. company search; read October 14, 2026.
17. Companies Registry, Hong Kong SAR. Registered Office Address of Live Local Companies (open data, JSON API). the portal opens the search of live local companies (its page title: Registered Office Address of Live Local Companies); the census queried its JSON API, data.cr.gov.hk/cr/api/api/v1/api_builder/json/local/search; read October 14, 2026.
18. Federal Trade Commission. ReportFraud. report form; read October 14, 2026.
19. National Association of Attorneys General. Find my AG. directory; read October 14, 2026.
20. U.S. Consumer Product Safety Commission. SaferProducts.gov. incident report form; read October 14, 2026.

T2 · The company's own pages and documents

1. Rad Life Mobility, Inc. (operator of radpowerbikes.com). Terms of Service, radpowerbikes.com. preamble naming the contracting company; page footer; read October 14, 2026.
2. Life Electric Vehicles Holdings, Inc. (OTC Markets news document). Life Electric Vehicles Holdings (OTCID: LFEV) CEO Issues 2026 Shareholder Letter. document id 92635; Rad Power Bikes acquisition completed February 27, 2026; QuietKat among its brands; issued August 24, 2026; read October 14, 2026.
3. Rad Power Bikes support site. Life EV completes Rad Power Bikes asset acquisition. company release; issued March 5, 2026; read October 14, 2026.
4. Yamaha Motor Corporation, U.S.A. (letter to dealers reproduced by CleanRider). Dealer letter ending U.S. e-bike wholesaling. undated letter to dealers; CleanRider's upload folder is 2024/11; read October 14, 2026.
5. myStromer AG (Stromer). Stromer FAQ for U.S. and Canadian customers. FAQ page; read October 14, 2026.
6. Salesforce. 2026 Holiday Predictions: A Unified Yet Fractured Journey (blog). the AI-assistant paragraph: '50% of shoppers report using an AI assistant at some point in their buying journey' and '74% of shoppers say they trust the product recommendations they receive from AI chat'; issued July 20, 2026; read October 16, 2026.
7. Salesforce. Shopping's New First Step: Agentic Search Grows 200% as Purchase Journeys Start in AI Chats (Fourth Edition State of Commerce report). 'Use of agentic search as the first step in the shopping journey grew 200% year over year'; surveys of 3,450 commerce professionals and 4,690 consumers (the methodology section prints 4,689); issued July 28, 2026; read October 16, 2026.
8. Clutch. Agentic Commerce: How AI Is Changing Online Shopping Behavior (survey of 403 US consumers, fielded December 17, 2025). Key findings: '65% of consumers have used AI assistants to research products before making a purchase, with 32% using them weekly'; Methodology; issued August 10, 2026 (the page's 'Updated' date; survey fielded December 17, 2025); read October 16, 2026.
9. Commerce, with PayPal (survey by Logica Research). Two-Thirds of Consumers Are Ready to Try Agentic Shopping, but Many Demand Human Approval Before AI Can Buy (press release). United States section: '67% interested in trying an agentic shopping tool'; 'consumers primarily want AI assistance during product research and comparison rather than during checkout. Purchase confirmation, fraud protection and strong payment security are the most important factors'; Methodology: 1,000 online shoppers in each of the United States, United Kingdom and Australia, March and April 2026; issued June 23, 2026; read October 16, 2026.
10. Adobe Digital Insights. Quarterly AI Traffic Report (April 2026). 'AI Conversion Now 42% Higher': 'In March 2026, they converted 42%' better (retail); 'Retail: +393% YoY' (October 2024 to March 2026); Methodology: consumer surveys of more than 5,000 U.S. respondents in March 2026; '39% of consumers have used AI assistants for online shopping'; issued April 2026; read October 16, 2026.
11. Shopify. agents.md.liquid (theme architecture documentation). 'Shopify generates an agents.md file by default for every store'; the /llms.txt and /llms-full.txt URLs serve the content of agents.md by default; undated (read October 16, 2026); read October 16, 2026.
12. Shopify. Agents: build agentic commerce experiences with the Universal Commerce Protocol (developer documentation). 'convert the cart into a checkout and refer them to the merchant storefront to complete payment. Trusted agents can complete checkouts directly.'; undated (read October 16, 2026); read October 16, 2026.
13. Universal Commerce Protocol (Google, Shopify and the protocol's contributors). Universal Commerce Protocol specification, version 2026-08-25: overview. the business profile at /.well-known/ucp ('A Business advertises its current protocol version and links to profiles for older supported versions'); issued August 25, 2026; read October 16, 2026.
14. Shopify. AI commerce at scale (news release). 'The Universal Commerce Protocol (UCP) is an open standard for AI agents to connect and transact with any merchant'; the AI channels named; issued January 11, 2026; read October 16, 2026.
15. llms.txt (Jeremy Howard's proposal site). The /llms.txt file: a proposal to standardise on using an /llms.txt file. 'A proposal to standardise on using an /llms.txt file'; undated (read October 16, 2026); read October 16, 2026.
16. Redtail eBikes. Shipping policy. Processing & delivery: 'orders are prepared and dispatched within 1–3 business days of confirmation'; 'most e-bikes arrive within 5–10 business days after dispatch'; free shipping within the contiguous United States; undated (read October 16, 2026); read October 16, 2026.
17. Redtail eBikes. Refund policy. damage: 'email us photos within 48 hours of delivery so we can resolve it'; warranty: 'covered under manufacturer warranty — we coordinate repair or replacement remotely'; undated (read October 16, 2026); read October 16, 2026.
18. Redtail eBikes. Contact page. 'We typically respond to email within one business day'; the phone line: 'Leave a voicemail if nobody picks up. Every voicemail is returned.'; undated (read October 16, 2026); read October 16, 2026.
19. Redtail eBikes (Shopify-generated). Universal Commerce Protocol profile of redtailebikes.com (/.well-known/ucp). version August 25, 2026; capabilities including dev.ucp.shopping.checkout, cart and buyer_consent; issued October 16, 2026 (read); read October 16, 2026.

T3 · Trade press, as corroboration

1. Bicycle Retailer and Industry News (BRAIN). Yamaha ending U.S. e-bike sales. news report; issued November 2, 2024; read October 14, 2026.
2. Bicycle Retailer and Industry News (BRAIN). Rad Life Mobility officially announces QuietKat purchase. news report; issued June 30, 2026; read October 14, 2026.
3. Bicycle Retailer and Industry News (BRAIN). Stromer ends sales in US and Canada. news report; issued November 20, 2024; read October 14, 2026.
4. Bicycle Retailer and Industry News (BRAIN). Domestic e-bike assembler EBC files for Chapter 7 bankruptcy. news report; issued October 8, 2025; read October 14, 2026.

T4 · Reference datasets and directories

1. National Association of Secretaries of State. Corporate registration: state business search directory. directory of state searches; read October 14, 2026.
2. Tranco. Tranco list N26LW (research-oriented ranking of the most visited websites). top 30,000, list of October 13, 2026; issued October 13, 2026; read October 14, 2026.
3. National Retail Federation (compiled by Kantar). Top 100 Retailers 2025. ranked list; issued 2025; read October 14, 2026.
4. Mozilla Foundation (Public Suffix List). Public Suffix List. list of October 7, 2026; issued October 7, 2026; read October 14, 2026.
5. Common Crawl Foundation. Common Crawl hyperlink graph, release cc-main-2026-jul-aug-sep, domain vertices. domain-vertices file, e-bike tokens; issued 2026 (release cc-main-2026-jul-aug-sep); read October 14, 2026.
6. Bike Index. Bike Index API v3 (manufacturer company URLs). manufacturers endpoint; read October 14, 2026.
7. Internet Archive. Wayback Machine. earliest capture; read October 14, 2026.

Image credit: the cover is a still life generated with Playcut (Nano Banana Pro) for this page; it shows no real place, person or product, and its type is set in real fonts.