Human Forward research finds 48% of assessed cycling websites meet its AI shopping criteria, while 55 brands receive no recommendations from eight assistants.

Cycling brands are ahead of major US retailers and brands on AI shopping readiness, according to research from Human Forward, but the study highlights gaps in how assistants recognise and recommend businesses across the sector.

The company tested 168 cycling brands operating in the UK and found that 70 of the 147 websites it could assess for AI shopping readiness met its criteria, equivalent to 48%. That compares with 16% in its separate test of 200 of America’s biggest retailers and brands.

Human Forward assessed readiness by checking for a Universal Commerce Protocol (UCP) manifest or storefront MCP endpoint. Its readiness figures record the presence of those standards, rather than a reported test of completed purchases.

Clothing brands led the categories, with 18 of 25 assessed sites meeting the criteria, including Castelli, Endura, Le Col and Proviz. Accessories followed at 13 of 20, including Knog, Exposure Lights, Muc-Off, Hiplok and Restrap. Bike manufacturers trailed at 19 of 55.

Five of 15 assessed retailer sites qualified: Sigma Sports, Condor Cycles, Cyclestore, Fully Charged and 99 Bikes. Bike manufacturers meeting the criteria included Pashley, Orange Bikes, Whyte, Temple Cycles, Cervélo and Colnago.

Several major names did not meet the study’s criteria, including Specialized, Brompton, Ribble, Giant and Canyon, alongside Halfords, Tredz and Merlin Cycles.

UK brands were ahead of international businesses in the assessed sample: 34 of 63 UK sites were classed as agent-ready (54%), compared with 36 of 84 international sites (43%).

Recognition does not always translate into recommendations

The research found a distinction between recognising a brand by name and recommending it within a cycling category. Oakley topped the study’s overall AI-answer scores at 99 out of 100, followed by Garmin at 98. Those scores are separate from website readiness: Oakley’s site could not be checked, while Garmin’s did not meet the readiness criteria.

Asked “What is SRAM?”, all eight assistants described computer memory. Yet every assistant recommended SRAM, Kryptonite and Hunt unprompted when asked for the best brands in their respective cycling categories.

Across the full sample, 33 of 168 brands were not correctly identified by any assistant when asked about their name alone. Other responses confused Ridgeback with the dog breed and 3T with an MRI scanner, while seven of eight assistants answered about a YouTube channel rather than Fully Charged, the London electric bike shop.

Meanwhile, 55 of the 168 brands were not recommended by any assistant when asked for the best brands of their kind in the UK. These included Colnago, Scott, Orbea, Ridley and Café du Cycliste, while Tweeks Cycles, JE James Cycles and Pedal On were among the bike shops not named.

Across the sample, 32 brands were recommended unprompted by all eight assistants. British brands in that group included Tredz, Brompton, Proviz, Sealskinz, Hunt, Brooks England, Muc-Off, Exposure Lights, Apidura and Hiplok.

The results underline the distinction between recommendation visibility and website readiness: Tredz and Brompton appeared in every assistant’s recommendations, but neither met Human Forward’s AI shopping criteria.

Community links remain limited

Human Forward also examined community links on the 152 home pages its crawler could read. Only 12 linked to rides, events or a members’ community run by the brand itself, including Rapha, Le Col, Pas Normal Studios, Café du Cycliste, Sigma Sports and Santa Cruz. The report says only five of the 12 ran a members’ community.

Thirty-five home pages linked to a Strava group, including Castelli, Ribble, Stolen Goat and Tredz. The check measured links visible on home pages, so brands may run rides or communities that were not captured.

Peter Abidi, founder of Human Forward, said: “Bike brands are better set up for AI shoppers than most of retail, and it’s the smaller ones leading the way. The gap now is recognition. If an assistant thinks SRAM is a memory chip and Kryptonite is something from Superman, your customer gets sent somewhere else. Community is the next gap. Strava is brilliant at bringing riders together, but very few brands give those riders a home of their own to come back to.”

How the research was conducted

The study was run on 6 November 2026 and covered 69 British and 99 international brands across bike shops, bike manufacturers, e-bikes, clothing, components, accessories and nutrition.

Five questions covering brand identity, where to buy, category recommendations, legitimacy and product ranges were put to ChatGPT, Claude, Perplexity, Google’s Gemini, Grok, DeepSeek, Meta’s Muse and Mistral’s Le Chat through their public APIs, with web search where offered. A separate AI model graded answers for brand mentions, official website links, accuracy and sentiment.

Website checks used HumanForwardBot, the company’s declared crawler, which follows robots.txt. Sixteen sites turned it away and five did not load. Human Forward stresses that blocking its crawler does not prove any named AI assistant is blocked, and that failing the readiness check does not mean a brand has no AI strategy. The results describe the sites and assistant responses tested on 6 November, rather than their ongoing status.

You can read the full report here.