Guide · October 2026

Does ChatGPT recommend your business?

The short answer: you can find out in about half an hour with a few honest prompts, and the reasons you are missing are usually fixable. AI assistants recommend businesses they can find, read and confirm in more than one place. This guide shows how they choose, how to test ChatGPT, Perplexity, Gemini and Copilot yourself, and what to fix first.

Covers
ChatGPT · Perplexity · Gemini · Copilot
Manual test
12 prompts, 3 scenarios
Checklist
12 points, on-site and off-site
Free tool
AI Visibility Audit, 0-100
The short versionAssistants recommend what they can find and verify: a readable website, consistent business details and other people talking about you. All three can be worked on.
No guarantees sold hereNobody can promise that ChatGPT will name your business, and anyone who does is guessing. What can be done is removing every reason for it not to.
Written by someone who fixes itI build websites and wrote the free AI audit that checks the technical half of the checklist below.

How AI assistants pick a recommendation

When someone asks an assistant "who is the best X in Y", the answer comes from two places. The first is what the model learned during training, a snapshot of the web that can be months or years old. The second, and for business questions often the one that matters, is a live web search the assistant runs while it answers. It sends one or more queries to a search index, reads a handful of pages and writes a summary with links to its sources.

Each assistant searches in a different way:

  • ChatGPT - uses its own crawler, OAI-SearchBot, to surface websites in ChatGPT search. According to OpenAI's help article, it also works with third-party search providers and rewrites your question into targeted queries for them; Microsoft Bing has been reported as one of them. Sites that block OAI-SearchBot are not shown in ChatGPT search answers.
  • Perplexity - builds its own index with PerplexityBot and shows numbered sources with its answers, which makes it the easiest assistant to audit by hand.
  • Gemini and Google AI Overviews - rely on the Google index, crawled by Googlebot. Google's models can also draw on Google Maps place data such as opening hours, ratings and reviews, and your Google Business Profile is where much of that data comes from.
  • Microsoft Copilot - Microsoft describes Copilot's answers as grounded in Bing web search results. A business that is weak or missing in Bing tends to be missing here too.

From the pages it retrieves, an assistant favours businesses that are named clearly, described consistently and confirmed by more than one source. In practice that comes down to a few kinds of signal:

  • Your own website - what you do, where, for whom and how to reach you, in plain HTML text that a crawler can read.
  • Reviews - on Google, Trustpilot, Tripadvisor, Clutch or whichever platform your industry uses. Assistants often summarise what reviewers say, so the words in a review matter, not only the star rating.
  • Mentions - articles, "best of" lists, local press, industry directories, association member lists and partner pages. Each one is an independent source saying that you exist and what you are good at.
  • Structured data - schema.org markup (Organization or LocalBusiness, Service, FAQPage) that states your name, address, services and contact details in a form a machine does not have to interpret.
  • Consistent NAP - the same Name, Address and Phone number everywhere. When your website, Google and three directories disagree about your address, a careful system has a reason to hesitate.
  • Business listings - Google Business Profile, Bing Places for Business and Apple Business Connect, which controls how a business appears in Apple Maps, Siri, Wallet and Mail. These are the structured sources that local answers lean on.

None of this is a secret ranking formula; the vendors do not publish one. It is simply what a system that summarises the web needs in order to say your name with confidence.

The manual test, step by step

Set aside about half an hour. You need the four assistants, a spreadsheet and a list of prompts written the way your customers talk. Do not start with "what is [your company]": real customers do not know your name yet.

  • 1. Prepare a clean session - use a private browser window or log out, and turn off chat memory or use a temporary chat, so earlier conversations about your business do not colour the answer.
  • 2. Make sure the assistant searches - in ChatGPT, check that the answer shows sources. If it does not, repeat the question with "search the web" added. Perplexity searches by default.
  • 3. Ask the category question first - who is good at X in Y, without your name. This is the test that matters.
  • 4. Then ask about you by name - to see what the assistant believes about your business and whether it is accurate.
  • 5. Ask each prompt two or three times - answers change between runs. One mention is luck; a pattern is a result.
  • 6. Record everything - for each prompt and assistant: the date, whether you were mentioned, your position in the list, which competitors appeared and which sources were cited.

Scenario A: a local business. Replace the brackets with your own service and city.

  • "What are the best [dental clinics] in [Manchester] for [implants]? Give me five with a short reason for each."
  • "I need [an emergency electrician] in [Leeds] today. Who should I call and why?"
  • "Which [hair salons] in [Wrocław] have the best reviews for [hair extensions]?"
  • "Is [Your Business Name] in [city] any good? What do customers say about it?"

Scenario B: a B2B service.

  • "Which companies in [the West Midlands] install [fire protection systems] for warehouses? I need a supplier with references."
  • "Recommend [an accounting firm] for [a 50-person software company] in [Munich] that works in English."
  • "Compare [Your Business Name] and [Competitor] for [commercial electrical installations]. Which would you choose and why?"
  • "Who are reliable [packaging suppliers] for [food brands] in [the Nordics]? Include their websites."

Scenario C: an international buyer. This is where many businesses never look, because the buyer is not in their city or even their country.

  • "Best immersive web design studio in Europe for a Dubai real estate launch."
  • "Which European studios build 3D websites for off-plan property developments? I need someone who can work with a team in Dubai."
  • "Find a [boutique hotel] in [the Algarve] that takes direct bookings and has [a spa]. Link to the official website."
  • "I am a buyer from [the UK]. Which [wineries] in [Croatia] ship to the UK and offer tastings in English?"

Then read the answers assistant by assistant:

  • ChatGPT - note whether the answer came from search (with links) or from memory (no links). If you appear only from memory, your details may be out of date. Ask a follow-up: "Which sources did you use?"
  • Perplexity - open the numbered sources. If competitors appear, the cited pages show exactly where their visibility comes from: their own site, a directory, a review platform or an article.
  • Gemini - compare what it says about your hours, address and services with your Google Business Profile. Errors here often start in the profile.
  • Copilot - if you are missing here but present elsewhere, check how your site looks in Bing. Verify it in Bing Webmaster Tools, which in 2026 added an AI Performance report (in preview) showing how often your pages are cited in Copilot and in Bing's AI answers.

The most useful part of the test is often not your own result but your competitors'. Every business an assistant names is there for a reason, and the cited sources show you what the reason is.

Technical causes: when AI cannot read you

If you are missing in every assistant, start with the website itself. These are the usual technical reasons, roughly in order of impact:

  • Content rendered only by JavaScript - Googlebot renders JavaScript; most AI crawlers do not. Vercel's analysis of AI crawler traffic (December 2024) found that the major AI crawlers it measured, including OpenAI's GPTBot and Anthropic's ClaudeBot, downloaded JavaScript files but did not execute them. A site built as a client-side app can look perfect in a browser and almost empty to them. Crawler behaviour changes over time, so test your own pages rather than relying on any single study.
  • robots.txt blocking the wrong bot - OpenAI, Anthropic and Perplexity each separate training crawlers from search crawlers. Blocking GPTBot keeps your content out of OpenAI's training data; blocking OAI-SearchBot removes you from ChatGPT search answers. Plenty of sites block both without knowing the difference. OpenAI's crawler documentation and Anthropic's help article list the current names.
  • Bot protection at the CDN or host - since July 2025, Cloudflare blocks known AI crawlers by default on newly added domains unless the owner allows them. Hosting firewalls and security plugins can do the same quietly.
  • A noindex left on - a tag or HTTP header from the staging phase that tells search engines to skip the page.
  • Key facts inside images or PDFs - a price list saved as a JPEG or a menu as a PDF is hard for a crawler to read and easy for an assistant to miss.
  • No structured data - the assistant has to infer who you are from loose text.
  • One page for everything - a single "Services" page listing twelve services gives an assistant nothing specific to cite. A separate page per service, written in the words your customers use, does.
  • Slow responses - crawlers work with time limits, and a page that takes many seconds to respond may simply not be read.

Off-site factors: what others say about you

A perfect website is still only one source, written by the party with the most to gain. Assistants look for confirmation elsewhere, which is why work outside your website matters at least as much as work on it.

  • Google Business Profile - complete categories, services, hours and photos, and replies to reviews. For local questions this is the listing to fix first.
  • Bing Places for Business - often neglected. It feeds Bing's local results, and Copilot answers are grounded in Bing.
  • Apple Business Connect - controls your place card in Apple Maps and how you appear in Siri, Wallet and Mail.
  • Reviews with substance - ask satisfied customers to mention the service and the place in their own words ("implants", "the Leeds office", "a Dubai launch"). Never buy or script reviews: platforms remove them, and consumer law in the UK and the EU explicitly bans fake reviews.
  • Mentions in independent sources - industry directories, trade associations, local press, partner pages, conference speaker lists and comparison articles. For B2B services, review platforms such as Clutch carry weight.
  • Consistent details everywhere - the same company name, address, phone number and website on every listing, social profile and directory. Old addresses and dead phone numbers create doubt, so clean them up.
  • A clear specialism - "web design" is a crowd; "immersive websites for off-plan property launches" is a category an assistant can match to a specific question. Being known for one thing makes you easier to recommend for it.

What about llms.txt?

llms.txt is a plain-text file at the root of a website, written in Markdown, that summarises who you are and links to your key pages. It was proposed in September 2024 by Jeremy Howard of Answer.AI, with a specification at llmstxt.org.

Be clear about its status: it is a proposal, not a standard. Google has said publicly that it does not use llms.txt, and I have not found a public confirmation from OpenAI, Anthropic or Perplexity that their assistants use it when deciding what to recommend. Its effect on recommendations is unproven.

I still add one to the sites I build, including my own, because it takes about fifteen minutes and costs nothing to maintain. Treat it as a tidy business card, not as a lever. If someone sells you llms.txt as the key to AI visibility, ask them for evidence.

The checklist: 12 points

  • 1. Search crawlers allowed - robots.txt does not block OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot or Bingbot.
  • 2. No bot wall at the CDN - firewall and CDN settings let verified AI crawlers through.
  • 3. No stray noindex - neither in the meta robots tag nor in the X-Robots-Tag header.
  • 4. Content in the HTML - your services, location and contact details are in the page source, not only rendered by JavaScript.
  • 5. One page per service - each with a clear heading in your customers' words.
  • 6. Structured data - Organization or LocalBusiness, Service and FAQPage in JSON-LD, matching the visible text.
  • 7. A sitemap - listed in robots.txt and kept up to date.
  • 8. An FAQ - the real questions customers ask, each answered in two or three sentences.
  • 9. Consistent NAP - identical name, address and phone number on the website and on every listing.
  • 10. Listings claimed - Google Business Profile, Bing Places for Business and Apple Business Connect, complete and current.
  • 11. Reviews with detail - a steady flow of genuine reviews that mention the service and the place.
  • 12. Independent mentions - directories, associations, press or partner pages that describe what you do.

Points 1 to 8 live on your website and can usually be fixed in days. Points 9 to 12 live elsewhere and take months of steady work. Start with the first group: it is the foundation the second group builds on.

Check it automatically

The technical half of the checklist can be tested in seconds. My free AI Visibility Audit fetches your page the way a crawler does, reads your robots.txt rules for 15 AI bot tokens, measures the content in the raw HTML, and looks for structured data, a sitemap and citable content. You get a score from 0 to 100 and a verdict for ChatGPT, Claude, Perplexity, Gemini and Copilot. Here is exactly what it checks.

It does not measure reviews, mentions or listings, and a high score does not guarantee a recommendation. It tells you whether the door is open. For the Google side, including speed, Core Web Vitals and the consistency of your business details, run the free SEO Audit as well.

When to get help

Most of the off-site list is work only you can do: asking for reviews, claiming listings, getting mentioned in your industry. The on-site part is a developer's job, and it is what I do.

I am Sebastian, and REELM is my one-person studio in Wrocław, Poland. I build fast business websites and immersive, cinematic websites with content in plain HTML, structured data and an open robots.txt as standard, and I fix these issues on existing sites. I work remotely in the CET time zone, which overlaps with the UK and the rest of Europe, sits two to three hours behind the UAE and leaves my afternoons for the mornings on the US East Coast. Every project is quoted individually after a 15-minute call.

Frequently asked questions

Can I pay to be recommended by ChatGPT?
No. ChatGPT has started showing ads to some users, but OpenAI says they are clearly labelled, shown separately from the answer and do not influence what ChatGPT recommends. What you can influence is how easy your business is to find, read and confirm.

How often should I repeat the manual test?
Once a month is enough for most businesses, plus after any big change: a new website, a new service page or a wave of new reviews. Use the same prompts each time so the results are comparable. Single answers move around, so look at the trend over several months.

Why does ChatGPT recommend my competitor and not me?
Look at the sources it cites. Usually the competitor has something you lack: a dedicated page for the exact service, more detailed reviews, a mention in a local article or directory, or simply a website the crawler can read. Each of those can be matched. You can also ask the assistant why it chose them, though its explanation is not always reliable.

Does llms.txt help ChatGPT recommend my business?
There is no public evidence that it does. llms.txt is a proposal from 2024, not a standard, and Google has said it does not use it. It is cheap to add and harmless, so I include one on the sites I build, but I would not expect it to change recommendations on its own.

Is good Google SEO enough for AI visibility?
It is a strong start, because Gemini and AI Overviews use the Google index and the same basics help everywhere. It is not the whole picture. Google renders JavaScript and most AI crawlers do not, Copilot relies on Bing, ChatGPT uses its own crawler and third-party search providers, and robots.txt rules can treat AI bots differently from Googlebot.

Make it easy for AI to say your name.

Run the free audit, then send me the link to your result. I will tell you which fixes matter for your business and what they would take. Every project is quoted individually.

Author: Sebastian. I run REELM, a one-person studio from Wrocław, Poland, and design and code immersive websites (WebGL, WebGPU, scroll-video) as well as fast business websites built to be read by search engines and AI assistants. Published: .