A client found this studio last month by asking an AI assistant for help with local SEO and an ecommerce store. Not by Googling. They asked a chatbot, it named a handful of studios, and they got in touch with one of them.
That is a referral route which did not meaningfully exist two years ago, and almost no small business is set up for it.
The Short Answer
- AI assistants mostly answer questions like “who should I hire for X in Y” by running a search and reading the results, not from memory. So if you don’t rank, you don’t get quoted.
- Being quotable is a separate skill from ranking: clear, factual, self-contained statements get extracted. Marketing waffle does not.
- The assistant has to be able to work out what you are, where you are and what you charge without guessing. That’s structured data, a consistent name and address everywhere, and specifics on the page.
- Check you aren’t blocking the crawlers. A surprising number of sites quietly forbid the exact bots that would otherwise cite them.
- Nobody can guarantee you a mention. Anyone selling that is guessing, and charging you for the guess.
Everything below is the detail behind those five points.
What This Is Even Called
Nobody has settled on a name yet. You’ll see AI SEO, GEO (generative engine optimisation), AEO (answer engine optimisation), LLM SEO and AI visibility, all describing roughly the same thing: being mentioned when somebody asks an AI assistant a question your business could answer.
The terminology will shake out eventually. The underlying work is the same whatever it ends up being called, which is worth knowing before you buy a package named after one of the acronyms.
How an Assistant Actually Decides Who to Mention
There are two routes, and the difference matters enormously.
From training data. The model absorbed a snapshot of the web at some point in the past. That knowledge is frozen, often many months old, and heavily weighted towards things mentioned repeatedly across many sources. You cannot realistically influence this in any reasonable timeframe, and for a small business you were probably never in there anyway.
From live retrieval. The assistant runs one or more searches, fetches the pages that come back, reads them, and writes an answer citing what it found. This is what happens for almost any question about a specific place, a current price, or who to hire.
That second route is the whole opportunity, and it has a blunt consequence:
If your business doesn’t appear in the search results the assistant retrieves, it cannot cite you. Classic SEO is not dead — it is the entry requirement.
The second requirement is being usable once retrieved. The assistant skims your page and needs to extract something concrete. A page that states “small business websites in Stirling typically cost £2,000–£5,000” gives it something to quote. A page that says “we deliver bespoke solutions tailored to your unique requirements” gives it nothing, and it moves on to a competitor who was clearer.
The Part Most People Miss: You Might Be Blocking Them
Many sites — often via a security plugin or a well-meaning setting — block the crawlers that feed AI assistants, then wonder why they never get mentioned.
The ones worth knowing about:
- GPTBot — OpenAI’s crawler
- OAI-SearchBot — the one behind ChatGPT’s search results
- ClaudeBot — Anthropic’s crawler
- PerplexityBot — Perplexity’s crawler
- Google-Extended — controls use of your content for Gemini
- CCBot — Common Crawl, which feeds a great many datasets
Check your robots.txt for any of those names. If they’re disallowed, you have opted out of exactly the thing you’re trying to appear in.
There is a genuine trade-off, and it deserves a straight answer rather than a recommendation dressed up as one. Blocking these bots reduces the chance of your content being used to train models you get nothing from. Allowing them is the price of being cited. For a business that wants enquiries, being cited is almost always worth more than the content protection. For a publisher whose product is the content, the sum is different.
One important nuance: Google-Extended does not control AI Overviews. Those are built from Google’s ordinary search index, so you cannot opt out of AI Overviews while remaining in Google Search. Anyone telling you otherwise is confusing two different things.
Bing Suddenly Matters Again
ChatGPT’s search has leaned heavily on Bing’s index, and OpenAI runs its own crawler alongside it. Perplexity and Copilot draw on Bing too.
Most UK small businesses have never once looked at Bing Webmaster Tools. If you aren’t properly indexed there, you’re invisible to a meaningful share of AI-assisted searches regardless of how well you do on Google.
It takes about twenty minutes to set up, it’s free, and it may be the highest-value neglected task on this entire page.
What Genuinely Moves It
Be a resolvable entity. The assistant needs to work out that “Fractal Studio” is one specific business in one specific place, not a phrase. That comes from structured data (LocalBusiness, Service, FAQPage), from your name, address and phone number being byte-identical everywhere they appear, and from third parties describing you the same way. Inconsistency here is the most common quiet failure.
Answer the question in the first paragraph. Then expand. The inverted-pyramid structure journalists use is almost perfectly suited to being quoted by a machine. If your answer is in paragraph nine, it will be skipped.
Publish specifics nobody else has. Real prices, real local knowledge, real numbers from work you’ve actually done. Generic advice is precisely what a language model can already generate without citing anyone — it only needs a source when the information is particular. This is the single biggest lever and the one most businesses refuse to pull, because publishing real prices feels exposing.
Keep classic rankings healthy. Retrieval pulls from search results. Everything in ordinary SEO — technical health, page speed, sensible structure, internal linking — still applies, because it decides whether you’re in the set of pages the assistant ever sees.
Get corroborated elsewhere. Reviews, directory listings, local press, third-party mentions. An assistant asked to recommend someone is far more comfortable naming a business several independent sources agree exists and is well regarded. This is also why Google Business Profile and review recency matter well beyond the map pack.
Make sure the content exists without JavaScript. If your text only appears after a script runs, some crawlers will never see it. Fetch your own page with JavaScript disabled and see what’s actually there.
llms.txt, with honest expectations. A plain-text summary of your site at /llms.txt, proposed as a convention in the same spirit as robots.txt. No major assistant has publicly confirmed using it as an input. It costs almost nothing, it’s a tidy summary of what you do, and it might matter later — this site has one. Treat it as cheap insurance, not as a strategy.
What Doesn’t Work
Guaranteed citations. Nobody controls what a model says. There is no submission form, no paid inclusion, no lever. A guarantee is the clearest available signal you’re talking to the wrong supplier.
“AI-optimised” content generated by AI. Publishing volumes of machine-written pages to appeal to machines produces exactly the generic text a model can already write. You become the least citable thing on the internet: unsourced, unspecific and indistinguishable.
Paying to be listed in an “AI index”. These are, so far, businesses selling directory entries with a new label on them.
Keyword stuffing for robots. Language models parse meaning, not keyword density. Writing for extraction means writing clearly, which is what you should be doing anyway.
Test Your Own Visibility in Ten Minutes
Worth doing before you pay anyone for any of this.
- Ask the assistants directly. Open ChatGPT, Claude, Perplexity and Google’s AI mode, and ask each the question a customer would: “Who are the best web designers in Stirling?” or “Recommend a WooCommerce developer in Scotland.” Note who gets named. Note whether it’s you.
- Ask about your business by name. “What do you know about [your business]?” This tells you what the models have resolved about you — and whether they’ve got it wrong, which happens more than you’d like.
- Vary the phrasing. People don’t type keywords at assistants, they describe situations. Try “my website is slow and I need someone local to fix it.”
- Check your
robots.txtfor the crawler names above. - Check Bing. Search
site:yourdomain.co.ukon Bing, not Google. If the results are thin, that’s your job for this week. - Validate your structured data. Run your homepage and one service page through a schema validator. Missing or broken markup is common and invisible.
- Re-read your own key page as a machine would. Does it state anything concrete and quotable in the first hundred words?
Write down the answers. Repeat monthly. That record is the only honest measurement available right now, and it’s more useful than any tool currently being sold for this.
What to Measure Instead
Expect your traffic numbers to look worse while your business does not.
More questions get answered on the results page, so fewer people click. The ones who do click have already read a summary, formed a view, and chosen to come to you anyway — so they arrive further along and convert better.
Measure enquiries, not sessions. A drop in traffic alongside steady or rising enquiries is not a problem to fix; it’s the shape of the change.
The Honest Caveat
This is early. The mechanisms are changing month to month, the terminology hasn’t settled, and anybody claiming a proven repeatable method is overstating what’s currently knowable — including anyone who writes a confident article about it.
What I’d stand behind is narrower: the work that makes you visible to AI assistants is almost entirely the work that makes your site genuinely good. Clear answers, real specifics, clean technical foundations, consistent identity, and enough third-party corroboration that you’re clearly a real business. None of that is wasted if the landscape shifts again, which is the main reason to do it.
If you’d like a straight assessment of where your site currently stands — including whether you’re accidentally blocking the crawlers, and what the assistants actually say when asked about your market — get in touch. A free review will tell you which of the above applies to you and which doesn’t.
