How trade businesses get recommended by AI search

Getting found24 August 202611 min read

Short answer: there is no AI-search switch. Assistants recommend what their sources support — crawled pages, business profiles, directories, reviews — and every platform that matters either documents how its system works or declines to. What a trade business actually controls is unglamorous: let the right crawlers in, keep the facts consistent everywhere, show real work, and test occasionally. Nothing here can be bought, and anyone promising otherwise is guessing with confidence.

The pitch to be sceptical of

Somewhere in your inbox there is probably an agency offering “GEO” — generative engine optimisation — at a few hundred pounds a month, promising ChatGPT will start recommending your firm. Google has a specific warning for this moment: its guidance on evaluating third-party SEO advice tells site owners to be wary of tools claiming special insight, and its AI documentation adds that no third-party tool has access to Google’s internal ranking or AI systems. Neither OpenAI nor Perplexity documents any such inside access for third parties either.

That does not make the underlying shift fake. People do ask assistants for a scaffolder now, and the answers draw on real web sources. It makes the sales pitch fake: certainty where none exists, priced monthly. The rest of this page is what the platforms document, minus the theatre.

How each surface actually works

“AI search” is four different systems wearing one label, and they do not work the same way. Treating them as one thing is the first mistake most advice makes.

Google (AI Overviews and AI Mode). Google’s generative-AI optimisation guide is explicit that these features are rooted in its core Search ranking and quality systems: pages are pulled from the ordinary index through retrieval-augmented generation, and the model runs related side-queries — query fan-out — to gather material. A page has to be indexed and eligible in ordinary Search before it can feature at all. The same page notes that a complete Business Profile can help a local business appear in AI responses, which quietly makes the Maps setup work part of the AI story too.

ChatGPT. OpenAI separates its crawling into documented roles. OAI-SearchBot reads the web for search: allow it in robots.txt and your site can surface in ChatGPT’s search features; block it and OpenAI states your site will not appear in those answers. GPTBot is a different crawler feeding model training, with its own independent setting — a business can reasonably allow one and decline the other. Robots.txt changes take roughly a day to register, per OpenAI’s crawler documentation.

Bing and Copilot. Microsoft runs its own crawler and its own business-listing product, Bing Places — same logic, different machinery: keep the listing accurate and the site crawlable by Bingbot. Firms that verify with Bing Places cover the surface most UK owners forget exists.

Perplexity. Answers come with visible citations drawn from the open web, and the company publishes its own crawler documentation. The practical posture is identical: crawlable, quotable pages that state clearly what you do, where you work, and what proves it.

What a trade business actually controls

Strip away the jargon and the controllable list is short, permanent, and mostly things worth doing regardless of AI:

• Crawlability. Public pages, sane robots.txt, no accidental blocks. This decides whether you are in the game at all.
• Consistent facts. Same name, service list, coverage area and phone number everywhere — site, Business Profile, directories, van. The three-surfaces model covers why matching facts across surfaces is the load-bearing wall of local discovery; assistants inherit whatever the sources agree on.
• Evidence. Photographs of real jobs, described honestly; services written as answers to the questions customers ask; areas served stated plainly. This is the raw material every one of these systems draws on.
• A site built to be quoted. Clear headings, direct sentences, one obvious subject per page — the same qualities a tired homeowner at 9pm needs. Nothing exotic; the full website guide covers the substance.
• Crawler choices made deliberately. For most trade businesses the reasonable posture is: allow OAI-SearchBot, decide on GPTBot knowingly, leave Googlebot well alone.
• Reputation left to accumulate honestly. Genuine reviews on the Profile and legitimate platforms. Purchased ones risk the listing everything else depends on.

Notice what is absent: secret factors, submission forms, monthly retainers. Google frames optimising for its AI features as ordinary SEO continued — and warns that mass-producing pages chasing every fan-out variant risks tripping its scaled content abuse policy. The boring version is also the safe one.

Test what AI search currently knows about your business

Ten minutes, once a quarter, free. Open two different assistants — say ChatGPT and Perplexity — and ask each the same five questions:

1. What does [your business] do?
2. Where is [your business] based?
3. Does [your business] serve [nearby town]?
4. What evidence is there that [your business] provides [main service]?
5. Who are reputable [trade] businesses in [area]?

Write down what comes back, especially anything wrong. Then fix the source. Wrong base town? Check the contact page and the Profile. Missing from the reputable-firm list? Look at who is listed and note where their public evidence lives — usually directories, reviews and cited pages, which is exactly the gap to close. Corrections reach the answers slowly; OpenAI notes even robots.txt changes take around a day, and content refreshes take longer still.

Two honest caveats. Answers vary between models, weeks, locations, search modes and whoever is asking — a different phrasing tomorrow may name different firms. And none of this is measurement in the way rank-tracking claims to be; it is a smell test. Its value is catching stale facts early, not producing a score to improve.

Measuring what can actually be measured

Google offers one genuine instrument: the Generative AI performance report in Search Console, covering impressions and clicks from its AI features. That leaves ChatGPT and Perplexity observable only from the outside — referral segments in analytics (chatgpt.com, perplexity.ai) and the quarterly prompt test above. Thin, but honest. Compare that with any dashboard claiming live AI rankings for £99 a month: both Google and OpenAI state outsiders have no access to their internal systems, so such numbers are reverse-engineered guesswork wearing a chart.

The chatbot on your site is not AI search

Worth its own paragraph, because the confusion is everywhere: an AI chat widget answers visitors who are already on your site. It has zero effect on whether ChatGPT, Perplexity, Copilot or Google’s AI features ever mention your business — different product, different direction of conversation. F12 ships one on every build because converting existing visitors pays for itself, not because it moves visibility. If an “AI SEO” package includes a chatbot as its main deliverable, the package does not do what its name suggests.

What nobody can honestly guarantee

Recommendations. Citations. Placement in an AI Overview. Not an agency, not a tool, not the platforms themselves — Google says plainly there is no way to pay for better positioning in its results, and since its AI features run on those same ranking systems, nothing can be bought there either. No competitor publishes a formula to optimise against even if you wanted to; OpenAI and Perplexity publish no internal metrics at all. Any guarantee on offer is either a misunderstanding or bait.

What remains is the deal every durable channel has always offered trades: be findable, be consistent, be provably good, and give it time. The surfaces change names every couple of years. That deal has not changed yet.

What F12 builds in, and how we check it

Everything above describes practice, not theory, so it is fair to show ours. Every F12 build ships with clean crawlability, structured data where it earns its place, an llms.txt file maintained as a courtesy to whichever machines read them — Google’s documentation confirms it neither helps nor harms Search visibility — and crawler permissions checked deliberately rather than inherited from a template. We run the five-prompt test on our own site too, and publish the results of what we find rather than screenshots of what we hope for. What we do not do is promise placements, quote internal metrics we cannot access, or bill monthly for a switch that does not exist. The published prices stay published for the same reason: the honest position is the whole position.

Questions people also ask