AI search is changing how shoppers find you. Here is what matters

Ask ChatGPT for "a waterproof hiking jacket under £150" and you will not get ten blue links. You will get three or four recommendations, written out in a paragraph, with a handful of sources credited underneath.

That is a different shop window, and a lot of retailers are trying to work out how to get into it. There is no shortage of people willing to sell them an answer.

So this edge-estion does two things. It explains what has genuinely changed, and it is honest about which parts of the current advice are worth your budget and which are not. Some of what is being marketed as essential right now has no measurable effect at all.

AEO and GEO: two acronyms, mostly one job

You will see both terms used, often in the same sentence.

• AEO (Answer Engine Optimisation): being the direct answer an assistant gives.
• GEO (Generative Engine Optimisation): being credited inside an AI generated summary, such as Google's AI Overviews.

In practice they overlap almost entirely, and plenty of good people use them interchangeably. GEO came from an academic paper in 2024 that used it as a catch all term, and the industry never tidied up after itself.

We are not going to pretend they are two separate services with two separate price lists. Underneath both labels sits one job: making sure AI tools can find your products, read the details correctly, and have some reason to mention you rather than the forty other shops selling the same thing.

What the evidence actually shows

This field has moved quickly enough that a lot of last year's confident advice has since been tested and found wanting. Here is where things stand.

What people say matters	What the research found
Your ordinary Google ranking	Still the strongest single predictor. Pages that rank well in normal search are far more likely to be cited in AI answers
Complete, accurate product data	Helps, meaningfully. Pages with full product details, including price, ratings and real specifications, were cited around 62% of the time, against 42% for pages carrying only basic information
Adding schema markup on its own	Little to no effect. Ahrefs tracked nearly 1,900 pages that added it and found no meaningful improvement in citations
llms.txt files and similar "AI ready" add ons	No effect. Google has said to skip them. No major AI provider uses them

Two things are worth taking from that table.

The first is that AI visibility is not a separate channel you can buy your way into. It sits on top of ordinary search performance. If your rankings are weak, no amount of AI specific tinkering will rescue you, and anyone suggesting otherwise is selling a shortcut that does not exist.

The second is that the thing which does help is not glamorous. It is having complete, accurate, well organised product information. Correct price. Real stock status. Genuine review scores. Proper specifications. It is the same unfashionable groundwork that has always mattered, and it is now doing double duty.

That is the good news for anyone already working with us, because it is largely the work we do anyway.

The technical issue behind most of the problems we find

There is a quirk in how modern e-commerce sites are built that catches out a lot of otherwise well run stores.

On many Magento setups, certain details, most often price and stock status, are not part of the page that first arrives from the server. They are filled in a fraction of a second later by the visitor's browser. There are sensible reasons for this. It is how Magento handles customer specific pricing, discount rules and personalised content.

Google's crawler waits for that to finish. Several AI crawlers do not. They take the first version of the page and move on.

This is worth keeping in proportion, because it tends to be described more dramatically than it deserves. Your page titles, your descriptions and most of your written content are usually present in that first version and come through perfectly well. What tends to go missing is the commercial detail: the price, whether it is in stock, the rating. Which is unfortunate, because when somebody asks an AI for "a good jacket under £150", the price is the entire question.

It is also a fairly contained fix in most cases. It is a change to how a few pieces of information are published, not a rebuild.

Working out whether it affects your store is not something you can check from the front end, which is why it forms part of the audit described below rather than something we would ask you to investigate yourself.

Your bot settings are worth a proper look

Two years ago, blocking AI bots was a popular reflex, and a fair few sites had it switched on without much thought about which bots were being caught.

It is worth revisiting, for two reasons.

The first is that "AI bots" are not one thing. OpenAI alone runs several crawlers doing different jobs. One collects data for training its models, which is a legitimate thing to have opinions about. Another builds the index that ChatGPT's search feature actually reads from. Block the second one and you are not protecting your content, you are removing yourself from ChatGPT's shopping results altogether. We have seen it happen by accident more than once.

The second is a date. From 15 September 2026, Cloudflare is changing its default bot rules, and the way the new categories interact means some sites that blocked AI bots in the past could end up blocking Google as well.

We would rather you did not go into your Cloudflare dashboard and start changing settings on the strength of a blog post, because the rules interact in ways that are easy to get wrong and the consequences of getting them wrong are severe. If your site sits behind Cloudflare, get in touch and we will review your configuration properly and tell you what, if anything, needs changing before the deadline. For clients on a support plan, this is something we can pick up as part of your existing hours.

What we can do for you

An AI visibility and technical audit

This is the piece of work we would suggest most stores start with. It covers three areas.

What AI tools currently say about you. You cannot look inside a language model and see what it thinks of your brand, and we will not claim to. What you can do is watch its behaviour, consistently and over time.

We use a specialist AI search monitoring platform to run a defined set of shopping questions, the ones your customers would realistically ask, across the engines that matter: ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Perplexity, Claude and Microsoft Copilot. Those questions are re-run every day, so what you get is a trend rather than a snapshot.

That tells us:

• How often your brand is mentioned, and how often your site is actually cited as a source, which are two different things
• Your share of the conversation against named competitors, and which questions you are winning or losing
• Which of your pages get picked up, and whether the details being quoted about you are correct
• How you are being described, not simply whether you are mentioned
• Which questions trigger shopping results, and whether your products appear in them
• Where third party sites, reviews and forums are being cited instead of you, which tells us where the gaps are

The questions are asked as a neutral user, with no account history and no personalisation, so what comes back reflects what a genuine prospective customer would see rather than what your own account would produce.

Choosing the right questions is most of the skill here. Thirty prompts that mirror how your customers actually shop are worth considerably more than three hundred generic ones, and that is the part we do rather than the software.

Whether your product data is reaching them. This is where we check the rendering issue described above, along with how complete your structured data is, whether it stays accurate when discount rules and customer group pricing apply, and whether configurable and bundle products are being described correctly rather than showing one misleading price.

Whether anything is blocking access. Crawler permissions, Cloudflare configuration, and anything else quietly getting in the way.

What you get back is a prioritised list. Not a hundred page document, but a short set of findings ordered by what will make a difference, with an honest note on effort against likely impact for each one.

Implementing the fixes

An audit that sits in an inbox is not much use to anybody, so the second half of this is the work itself.

Depending on what turns up, that usually means correcting and extending your structured data so it is complete and stays accurate as the catalogue changes, moving key commercial information into the part of the page that every crawler can read, sorting out crawler and Cloudflare access, and improving the depth of product information on your highest value pages.

For clients on a support plan, most of this can be scheduled into your existing hours rather than quoted separately. Worth asking us about at your next check in.

One thing we will not do is promise you a place in AI answers. Nobody can sell you that, and we would be wary of anyone who tries. What we can do is make sure that when an AI tool or a search engine comes looking, it finds correct and complete information rather than gaps, and that nothing is standing in its way.

Edge-estions: things worth doing yourself this month

Not everything here needs us. Here are four things you can usefully do without picking up the phone.

1. Ask an AI about your own products, then treat the answer with some caution. Open ChatGPT or Perplexity and ask the questions your customers would ask. "Best [your product category] under £X." "Where can I buy [your best seller] in the UK?" See whether you come up, and if you do, check the price and stock information it quotes. Getting this wrong is more common than being left out altogether, and arguably worse.

One thing to know before you start. The answer you get is personalised to you. These tools remember your previous conversations, have a rough idea where you are, and adjust accordingly. Ask about your own brand from your own account and you are one of the most likely people in the country to be shown it. That is exactly why our reporting asks the same questions as a neutral user with no history behind it. If what you see does not look right, or looks noticeably better than the figures in your report, send us the screenshot and we will work out why.

2. Fill in the dull fields. Product codes, brand, materials, dimensions, delivery costs, returns policy. Tedious, and it is the single thing that correlates most reliably with getting recommended. Your team can work through this without any developer involvement.

3. Write something the manufacturer did not. If your product copy is the supplier's PDF, you are one of forty identical pages and there is no particular reason for an AI to quote you rather than any of the others. Sizing notes based on what actually gets returned. Who the product does not suit. Your own photographs. That is the material that makes a page worth citing.

4. Keep going with the ordinary search work. Position in normal search results is still the best evidenced predictor of whether you get mentioned in an AI answer. If you have been wondering whether your existing SEO budget is still the right call now that everything has changed, the evidence says it is.

The wrap up

Less has changed than the noise suggests, and the parts that have changed reward the same things good retailers were doing anyway: accurate product information, content that is not a copy of everybody else's, and a site that is straightforward to read.

What is genuinely new is that the cost of getting the basics wrong has gone up. A missing price used to be a minor annoyance. Now it can be the reason you are left out of a recommendation entirely.

If you are already on a support plan with us, the simplest next step is to mention this at your next check in and we will get an audit scheduled. If you are not, get in touch and we will take a look at how your store is currently being read, what is being said about you, and what is worth fixing first.

Frequently asked questions

Is this replacing SEO, or adding to it?

Adding to it, and largely built on it. Ranking well in ordinary search remains the strongest predictor of whether AI tools cite you. Treat AI visibility as an extension of your search work rather than a replacement for it.

Will AI Overviews stop customers visiting my store?

They will reduce your visitor numbers. Zero click searches in the US reached 68% in 2026, up from 60% in 2024, and Ahrefs found that click through at the top result falls by around 58% when an AI Overview appears above it.

The e-commerce picture is more encouraging than that sounds. A 2026 study across 100 e-commerce sites and roughly 33.5 million sessions found that while organic visits fell after AI Overviews launched, conversion rates went up. Fewer people arriving, but more of them ready to buy. Plan for lower traffic at higher intent rather than assuming the numbers hold.

Why does your report say something different to what I see when I ask ChatGPT myself?

Because you are not a neutral user. These assistants remember your previous conversations, factor in roughly where you are, and personalise what they show you. Asking about your own brand from your own account is close to a best case scenario, and it is not what your customers experience.

Our monitoring deliberately asks each engine as a fresh user with no history, from a consistent location. The figures tend to look lower as a result, and they are considerably more useful. If you spot a difference that seems bigger than that explains, send it over and we will look into it.

Is this a one off report or something ongoing?

Both, and ongoing is usually the more useful of the two. The initial audit tells you where you stand and what to fix. After that the questions are re-run daily, so you can see whether the changes we made moved anything, and pick up on it when a competitor starts appearing in answers where you used to. For most support plan clients we fold this into regular reporting rather than quoting it separately.

Does site speed affect whether AI tools recommend me?

Not directly, as far as anybody can measure. There is no evidence that ChatGPT or Perplexity assess how quickly your site loads. It matters indirectly, because speed feeds into your Google ranking, and your Google ranking feeds into AI citations. Worth doing, just not for the reason it is usually sold.

Do shoppers still need to visit my site to buy?

Less than they used to. ChatGPT now supports completing a purchase without leaving the conversation, although it is US only at the moment and the systems for merchants to take part are still being built. For UK retailers this is something to keep an eye on rather than plan around this quarter. But the assumption that a sale requires a visit to your website has a shelf life.

Should we add an llms.txt file to our site?

No. It has been marketed fairly hard, Google has publicly said to skip it, and no major AI provider uses it. If it appears on a proposal you have been sent, that is a reasonable thing to ask about.

How quickly would we see a difference?

Fixing information that is not reaching AI tools shows up quickly, because it is the difference between readable and not. Beyond that, gains track your ordinary search rankings, which move on normal SEO timescales, so months rather than weeks. Anyone quoting you a fixed number of weeks for AI citations is guessing.