Why your conversion rate looks worse than it is going into Black Friday

On one store we work with, half of October's sessions weren't people. Here is what that does to a brand's numbers, its email list and its place in AI answers going into peak.

In October, 51% of the sessions on a Shopify store we work with didn't come from a person. That is roughly half a million sessions in a single month. In October last year, the figure was 5%.

Our Head of Ecom Strategy flagged the number to the team this week. The traffic isn't buying anything. It is crawling. His read is that these bots are most likely learning, gathering what they can about the store's products, and he can't see why anything would need half a million sessions a month just to check for updates. His caveat is that the picture will vary from site to site. Even so, a jump from one session in twenty to one in two, in twelve months, is not something any brand should find out about after Black Friday.

It also landed in the same fortnight that Google and OpenAI each made it easier for shoppers to find and buy products without visiting a brand's website at all. The two stories are connected, and together they change what a brand's website is for.

What changed this month
‍

On 4 November, Google switched on direct checkout inside AI Mode in Google Search and the Gemini app for shoppers in Canada. It runs on the Universal Commerce Protocol, an open standard Google launched earlier this year with industry partners so that AI agents can complete a purchase on a shopper's behalf. The first retailers include lululemon, select Shopify merchants, Shoppers Drug Mart and Simons. A shopper researching a product in AI Mode can now buy it in the same conversation, and the retailer stays the seller of record. Google plans a simpler set-up in Merchant Center next year so more retailers can switch it on themselves, and has said the rollout will reach the UK after Canada and Australia.

Six days later, OpenAI launched Businesses in ChatGPT in beta. Brands can claim a verified profile, keep their details, prices and stock up to date, upload product feeds and, for the first time, see data from ChatGPT itself on how often it mentions them. Profiles can be claimed in any market where ChatGPT is available, and Shopify merchants can connect their store directly. Eligible US ecommerce businesses can also launch a Business Agent, a branded chat that answers shoppers' questions using the brand's own product information. That agent can also be sponsored as the destination for ChatGPT Ads.

Neither of these is live for UK checkout yet. Lucy Davis, our CCO, who leads our work on AI search, has been clear that it is only a matter of time before Google's checkout reaches the UK and Europe. Brands selling into the US and Canada are already inside it.

Why the bot share matters right now
‍

A large share of non-human traffic would be a reporting nuisance in any month. Going into the busiest weeks of the year, it becomes a commercial problem.

The first issue is data. As our Head of Ecom Strategy puts it, getting reliable numbers is very messy when more than half the traffic is bots. Every session that never had a chance of buying still counts as a visit, so conversion rates look worse than they are and traffic looks healthier than it is. Shopify's reports have filters that separate human and bot sessions, and many brands rely on them. On this store, the filter excluded most of the bots in October, but in earlier months the results were mixed, and at times it removed only 5 to 10% of the bot traffic we had identified. A filter that works in some months can't be relied on going into peak. Peak is also when brands watch their on-site funnel most closely, from add-to-cart rates to how many shoppers begin checkout. A sudden drop in those numbers can raise a red flag and send a team hunting for a bug that isn't there. Those numbers rest on a denominator that is, on this store, half machine. Paid channel reporting is less exposed, because spend and results are measured inside the ad platforms.

The second issue is less obvious. On the same store, the bots were triggering CRM flows, mostly abandoned cart emails, which send automatically. Visits that aren't real people were setting off automated emails to fake addresses. That pollutes the customer data, and it also damages domain health, because sending to addresses that bounce or never open tells inbox providers the sender isn't being careful. Black Friday is the week when a brand most needs its emails to land in the inbox rather than in spam. Our Head of Ecom Strategy has since excluded the bots from the store's flows. Standard filters didn't catch them, because the bots don't behave like normal visitors, so he built the exclusion around actions that every real shopper takes and the bots never do.

Why blocking everything would be the wrong fix
‍

The obvious response is to shut the bots out. For most brands, that would trade one problem for a bigger one, because some of those crawlers are how a brand gets into AI answers in the first place.

We covered this in October, after Lucy spent two days at BrightonSEO. One of the clearest examples came from Adam Preston of Summit: a UK pharmacy retailer, Weldricks, with 232,000 organic visits a month and strong Google rankings had an AI visibility score of 0.0. Greta Koivikko of NOVOS showed that most AI crawlers can't read content that loads with JavaScript after the page opens, so when an assistant is asked a question about a product, it answers from reviews and other people's websites instead. A brand that blocks crawlers wholesale would be choosing that outcome on purpose.

The launches from Google and OpenAI make this sharper. Both run on product data that machines read directly, whether that is a Merchant Center feed or a feed uploaded to ChatGPT's Business Hub. The job for a brand is to separate the machines that help it get found from the ones that only add noise and cost, and to stop the noise from reaching its reporting and its CRM. Cloudflare works with Shopify for managing this kind of traffic, and it is one of the options our Head of Ecom Strategy is looking at for the brands we support.

What to check before peak
‍

None of this needs a rebuild before Black Friday. It does need someone to look. These are the checks we would start with, drawn from what we are seeing on a client store and from our October work on AI search.

  1. Know the share. Find out how much of the store's traffic is non-human now, compare it with the same month last year, and check whether the platform's bot filter caught it in each of the last few months. On the store above, that comparison was what turned a vague suspicion into a clear problem.
  2. Read conversion in that light. If the bot share has jumped, conversion rate and traffic trends from the last few months need to be read with that in mind before a dip in add-to-cart or checkout rates is treated as a bug on the site.
  3. Check which flows bots are triggering. On this store it was mostly abandoned cart emails, which send automatically. The usual filters may not catch the bots. What worked here was excluding visitors who skip an action every real shopper takes. A sunset flow, which stops emailing addresses that never open, is part of the retention set-ups we have built for clients because it protects deliverability for everyone else on the list.
  4. Claim the brand's ChatGPT profile. It is free during the beta and open wherever ChatGPT is available, and it is the first time brands can see ChatGPT's own data on how often they are mentioned.
  5. Make product data readable. Stijn Bergmans of Follo told BrightonSEO that most feeds he audits are only around 60% complete, with the optional fields that help a product get recommended left empty. Key product details should sit in the page's source code rather than loading later. And since ChatGPT's US shopping index cuts descriptions off after roughly 40 words, the most important terms belong at the start.

On measurement, Stijn also showed how little of AI's influence appears in standard analytics: across his agency's clients, AI tools account for 0 to 1% of tracked traffic, yet roughly half of clients say they first came across the agency in an AI conversation. Branded search and direct visits are better signals of that influence than a referral report. With half the sessions on some stores now coming from machines, the gap between what analytics shows and what is actually happening is only getting wider.

Where judgement still has to sit with a person
‍

No tool will decide which bots a brand should welcome. That call depends on where the brand sells, how much it relies on organic and AI discovery, and how much damage the bad traffic is doing to its data and deliverability. A brand selling heavily into the US, where both Google's checkout and ChatGPT's Business Agents are live, will weigh it differently from one that sells only in the UK today. The figure from our own client comes with the warning that it will look different on other sites, and the right response starts with checking a brand's own numbers, not with copying someone else's settings.

It also crosses teams. The bot problem shows up in analytics, in the CRM and in the product feed, and each of those usually has a different owner. As Lucy put it after BrightonSEO, "SEO used to be the responsibility of one person or a team, and now AEO is a cross-functional strategy that touches all teams in different ways." Bot traffic is the same kind of problem, and it tends to go unnoticed for the same reason.

What this means for a brand's website
‍

For years, a brand's website was built for one kind of visitor. On the store we started with, half of last month's visitors were machines reading the catalogue, and the AI assistants they feed are starting to take the sale without the shopper ever landing on the site. Through Offcuts we have been talking about this shift for a while, from what AI search means for brands built on culture to how agentic storefronts are changing conversion. This month it stopped being a forecast and showed up in a client's analytics.

Brands now have to be readable by the machines that matter and protected from the ones that don't, and they have to make that call before Black Friday rather than after it. A year ago, one session in twenty on that store was a bot. Now it is one in two.

We work on AI search with brands now, including as CGC's AEO partner. If you want to know how AI crawlers and assistants are treating your store before peak, get in touch.

‍

What we’re solving in commerce this week

Each week, a real problem from inside a brand we work with and how we solved it.

Read by people at Stripe, Shopify, Dr. Martens, Loewe, H&M and Mercedes-AMG F1, along with hundreds of other brands shaping ecommerce.