How to get recommended by ChatGPT: the checklist we run for clients

Short answer: you get recommended by ChatGPT the same way you get found on Google — by being easy to find, easy to read and easy to trust. When someone asks an assistant for a local recommendation, it runs a live web search, reads a handful of pages and cites the ones that answered the question. If your pages can't be retrieved, or they don't state the facts plainly, you're not in the draw.

Most 'AI visibility' advice stops at 'create great content and hope'. What follows is the checklist we actually run for clients: confirm the crawlers can reach you, put your facts on the pages assistants like to retrieve, mark them up so machines can lift them cleanly, then test with real buying questions and log who gets cited. One caveat up front: you can influence citations, not control them — anyone promising guaranteed placement is selling snake oil.

How ChatGPT and other assistants actually pick who to recommend

Assistants don't 'know' your business the way Google knows it. As of October 2026, there's no ChatGPT-specific profile to claim and no score to optimise — though Bing Places feeds the Bing index ChatGPT's search leans on, so claim that (step 1 covers it). When someone asks for a local recommendation, the assistant runs a live web search, reads a handful of pages and cites the ones it used. No retrieval, no recommendation.

The plumbing differs by assistant, which is why we test in more than one:

• ChatGPT leans on Bing's index plus its own crawler, OAI-SearchBot.
• Perplexity sends PerplexityBot and supplements with other indexes.
• Gemini grounds its answers in Google Search.

Same job each time: retrieve pages, judge which ones answer the question. A recommendation is retrieval plus judgement — and you can work on both.

It's the same first mile Google runs — crawl, index, retrieve — just with different machinery. More on crawling and indexing in our guides.

One nuance worth a sentence: for general knowledge questions, assistants can answer from memory. Local questions and 'who's good near me' questions trigger live search. That's the opening for a small business — the assistant has to go looking, and you can influence what it finds.

Step 1: Make sure the AI crawlers can reach your site

Before anything else, check the crawlers can actually reach your site. Most businesses have never looked, and some are accidentally blocking the very crawlers that would cite them.

Open yourdomain.co.uk/robots.txt in a browser and search the page for five names: GPTBot, OAI-SearchBot, PerplexityBot, Google-Extended and ClaudeBot. A block looks like this:

User-agent: OAI-SearchBot
Disallow: /

Some hosting platforms and security plugins now ship with a 'block AI bots' toggle switched on by default. Cloudflare made this the default for new domains in 2025, and it's easy to switch on without noticing — so check rather than assume.

Know which crawler a toggle blocks before you leave it on. They do different jobs:

• Training crawlers (GPTBot, ClaudeBot) learn from your content. Blocking them stops your text training future models.
• Search crawlers (OAI-SearchBot, PerplexityBot) retrieve your pages live when someone asks a question. Blocking these removes you from the answers entirely.

Blocking training crawlers is a legitimate choice. Blocking search crawlers while wondering why ChatGPT never recommends you is not — whichever way you go, make it deliberately.

Next, Bing. Most small businesses have never opened Bing Webmaster Tools; verification takes a few minutes, and the pages report shows what Bing has. Because ChatGPT's search leans on Bing, this matters. Our free SEO audit flags indexing and technical problems generally; for Bing specifically, open Bing Webmaster Tools and look at the pages report. While you're there, claim your Bing Places listing — it feeds the Bing index that ChatGPT's search leans on, and it's worth doing.

Finally, the pages you most want found must be public: no login walls, no password-protected staging site accidentally left live, no 'coming soon' sitting on the service URL that matters most. All three turn up regularly on small-business sites.

The readiness check below answers in under a minute; the full step — robots.txt, Bing, public pages — is a 20–30 minute job the first time through.

Step 2: Put your facts on the pages assistants like to retrieve

In our experience, most citations come from three places: the business's own service pages, its Google Business Profile, and the review or directory sites in its trade. Assistants like aggregators because one page answers 'who's good near me'; a good directory listing does in one fetch what would otherwise take ten. So directory presence isn't old-school SEO clutter. It's retrieval surface.

Your own site. One page per service, written plainly: what you do, where you do it, prices or honest ranges, hours, qualifications, who the people are. A page called 'Boiler repair in Reading' with the actual details gives the model something to lift. Vague 'solutions provider' copy gives it nothing to quote.

Your Google Business Profile. Complete every field, correct categories, services listed, photos, a steady flow of reviews. If you do only three things: get the primary category right, list every service with a sentence on each, and get reviews moving that name the job and the town. For trades and clinics this is often the single biggest lever in the checklist. There's more on local SEO in our guides; if you'd rather hand it over entirely, local SEO and Business Profile management is standard work for us.

The review sites in your trade. Checkatrade and Trustpilot for trades, CQC profiles for clinics in England, TripAdvisor for hospitality. Check which directories your competitors are on and be there too. If competitors are listed somewhere you're not, that's a gap to close — the assistant can only cite what it can retrieve.

Reviews do double duty. Assistants summarise review text, so a review that names the service and the town — 'replaced our boiler in Reading in a day' — is worth more to a citation than 'great service'. Worth mentioning next time you ask a customer for one.

Step 3: Mark up your details so machines can lift them cleanly

Schema is machine-readable labelling: it says what a fact is and whose it is, in a format machines parse without guessing. Add three things:

• LocalBusiness schema — or the specific type, such as Dentist, Plumber or Restaurant — on your homepage or contact page
• Service schema on each service page
• FAQPage schema for the questions customers actually ask (Google restricts FAQ rich results in its own listings to government and health sites now, but the markup still hands an assistant a clean list of questions and answers)

Then the unglamorous part: keep your name, address, phone, hours and service list identical across your site, your Business Profile and every directory you're on. Contradictions make the model hedge, and hedging models don't recommend — they say 'you could try…' and move on.

Two honest limits. Schema won't force a citation — what it does is remove ambiguity about which fact belongs to whom, the same job it does for Google's rich results. And llms.txt, a proposed file some sites publish for AI crawlers: there's no solid evidence assistants use it yet. We don't charge clients for it and don't pretend otherwise.

Step 4: Test like a customer — and keep a log

This is the step almost nobody else publishes, and it's the one that turns the rest from theory into evidence.

Write 10–15 questions phrased the way customers actually ask. Always place plus trade, never just 'best plumber':

• 'who's a good emergency plumber in Reading'
• 'best physio in Leeds for shoulder pain'
• 'family dentist in Bristol taking NHS patients'

Run each question in ChatGPT, Perplexity and Gemini — in a fresh chat every time, so one answer doesn't colour the next. Note who gets cited and, more importantly, which page or source each citation came from. Most assistants show their sources; click through.

Log five things per test:

1. Date
2. Question
3. Assistant
4. Businesses cited
5. Source of each citation

A spreadsheet is fine. Ours has five columns — date, question, assistant, who was cited, and the source page — and takes about a minute per question.

Repeat monthly. Citations move between runs and between model updates — a single test tells you almost nothing; a trend tells you plenty.

Without a log, you're guessing whether anything you changed did anything. This is the difference between AI visibility work and AI visibility superstition.

Step 5: Read the log and close the gap

A log only earns its keep if you act on it. Three patterns cover most of what you'll see.

A competitor is cited and you're not. Find where the citation came from — ask the assistant which source it used, or click through. Usually it's a directory, a review site or one specific service page. Then close the gap in this order:

1. Get on the missing directory or review site
2. Deepen the matching service page on your own site — more specifics, not more adjectives
3. Fill the empty Business Profile field
4. Work on reviews that name the service and the area

You're cited but described wrongly — wrong services, wrong area, wrong opening hours. That's a facts problem. Fix the source page, not the prompt. Rewording your test questions until you like the answer measures your prompting, not your visibility.

Nobody is cited consistently. Either the question is too vague for anyone to win, or the assistant genuinely has thin sources for that trade in that town — which, if it's your town, is an opportunity.

Expect fixes to take weeks to show up in tests, not days. Retrieval indexes refresh on their own schedule, and no amount of re-testing speeds that up.

What you cannot control — and how to spot snake oil

The honest position: you can influence citations, you cannot control them. The same question can return different answers in the same week. As of October 2026, there is no paid placement to buy and no trick that forces a citation — so anyone promising guaranteed ChatGPT placement is selling snake oil. We turn that work down, and you should too.

Model updates reshuffle results. Whatever works today needs re-testing after the next update, which is exactly why the monthly log matters more than any one-off 'optimisation'. A log survives an update; a promise doesn't.

This isn't a replacement for SEO, either. Crawlable pages, clear facts, real reviews and a complete Business Profile are the same fundamentals Google rewards — assistants just read them differently. Do the basics well and you're working on both at once.

One warning about tools: be wary of AI visibility dashboards that never tell you which source page a citation came from. The source is the actionable part — it tells you what to fix. Everything else on those dashboards is theatre.

Your first hour: what to do today

The timings below are honest ones — the list fits inside an hour unless your Profile needs serious work:

1. Run the AI readiness check — it covers crawler access and structured data in one pass, and answers in under a minute.
2. Open your robots.txt and search it for GPTBot, OAI-SearchBot and PerplexityBot. Ten minutes. If any are blocked, decide deliberately whether that's what you want.
3. Audit your Business Profile: primary category, full service list, review flow. Twenty minutes if it's in decent shape.
4. Write your first five test questions and run each once in ChatGPT, Perplexity and Gemini. Fifteen minutes. Save the results — that's month zero of your log.
5. If the check turns up problems you'd rather not fix yourself, book a free call. This checklist is literally what we run for clients, so there's no hard sell needed.

We'd rather you did the first four steps yourself and called us only if the log shows a gap you can't close. A business that tests monthly gets more out of any help than one that outsources and never looks.

Questions people ask

Found this useful? Share on LinkedIn · Email it to someone