Getting your business recommended by ChatGPT and Claude: what actually works

There is no ranking position to chase. As of 2026, how to rank in ChatGPT and Claude results comes down to being mentioned or cited in the answers they generate. That depends on three things: having clear, factual, crawlable pages that answer specific questions; being named consistently across trusted third-party sites; and allowing AI crawlers to access your content. No one can guarantee a particular mention.

Most small UK businesses ask this question as if ChatGPT and Claude had a page one. They don't. Each answer is generated fresh, and the same question asked twice can produce different brands. That changes what you should aim for: not a position, but a higher probability of being included.

How ChatGPT and Claude decide which businesses to mention

These tools draw on two sources. The first is what the model learned in training, which is a fixed snapshot of text and is not something you can edit. The second is live retrieval: when a tool searches the web to answer a question, it pulls in pages and summarises them, usually with citations. Whether a given answer uses retrieval depends on the product, the settings and the question.

The practical consequence is that you influence the second source far more than the first. Pages that are indexed, fast to load, and written so a machine can lift a clean answer from them are the ones that get pulled in. Pages hidden behind scripts, login walls or vague marketing copy rarely are.

What the models tend to favour

Neither OpenAI nor Anthropic publishes a ranking formula, so anyone claiming to know the exact weighting is guessing. What is observable, and consistent with how retrieval works, is that cited pages tend to:

• answer a specific question in the first few lines;
• state facts plainly (prices, locations, service areas, qualifications);
• appear on sites that other sites mention or link to;
• have clear structure, with headings that match how people phrase questions.

Google's own Search Central documentation says that helpful, people-first content and good technical foundations matter for search. Those same foundations are what most retrieval-based tools lean on, because many draw from search indexes. Strong traditional SEO is therefore the best starting point rather than a separate project. If you are short on resources, our piece on improving website visibility with no SEO team covers the basics.

The technical and reputational details that trip people up

Most failures here are mundane, not strategic.

Crawler access

OpenAI documents its crawlers (including GPTBot and OAI-SearchBot), and Anthropic documents ClaudeBot. You control access through your robots.txt file. Some website platforms and security plugins block AI crawlers by default, so a business can be invisible without ever choosing to be. Check your robots.txt and any bot-protection settings. Note that blocking a training crawler and blocking a search crawler are separate decisions, so read each provider's current documentation before changing anything.

Consistency of your business details

If your name, address, services and descriptions differ between your website, Google Business Profile, Companies House record and directory listings, the models have conflicting information to summarise. Fix the discrepancies first. For UK businesses, a consistent trading name and a clear statement of the areas you serve matter more than most owners expect.

Third-party mentions

Models often reflect what others say about you. Trade body listings, reputable local press, industry roundups and genuine reviews all feed that picture. Self-published claims carry less weight than independent confirmation, so a page saying you are "the best in the North West" does little by itself.

Content that can be lifted

Write the answer first, then the detail. A page on "how much does a boiler service cost" should state a realistic UK range near the top, then explain what changes the price. Where you use AI to draft this, keep it factual and in your own voice; our guide on what separates good content automation from a Google penalty explains the quality line.

The same business, two approaches to AI visibility

This is an illustration on hypothetical figures, not a real business. Assume a small accountancy practice in Leeds. Assume it tests 20 realistic customer questions (such as "who can help a sole trader with Making Tax Digital in Leeds?") in ChatGPT and Claude, and that in the baseline test it is mentioned in 2 of the 20 answers.

Approach A: publish more, change nothing else. The practice adds 10 blog posts of general tax commentary. Pages are long and vague, robots.txt still blocks AI crawlers through a plugin default, and business details differ across listings. Assume a retest shows 3 mentions in 20. That is 3 ÷ 20 = 15% of answers, up from 2 ÷ 20 = 10%.

Approach B: fix foundations, then publish fewer, sharper pages. The practice unblocks the relevant crawlers, aligns its listings, and writes 4 pages that each open with a direct answer and clear local service details. Assume a retest shows 7 mentions in 20, which is 7 ÷ 20 = 35%.

Both figures are assumptions chosen to show the logic; real results vary by sector, competition and the day you test. The point is that the single changed variable, foundations plus focused pages versus volume alone, is what plausibly drives the gap. Because answers vary run to run, repeat each test several times before drawing conclusions.

Where the guidance runs out: what no one can promise about AI answers

Be sceptical of anyone selling guaranteed placement.

• Answers are probabilistic. The same prompt can name different businesses at different times, so any single screenshot proves little.
• Training data is fixed. Changes you make today will not alter what a model already learned. Only retrieval-based answers can reflect new content quickly, and not every answer uses retrieval.
• No official ranking factors exist. OpenAI and Anthropic do not publish a checklist. Advice here, including this article's, is informed inference from how the systems behave and from search documentation.
• Paying does not buy a place. Neither tool lets you pay to be recommended in an organic answer.
• Some queries will never mention you. Where a question has a single dominant authority, such as a regulator or a national brand, a small local firm may not surface however good its content is.
• Measurement is imperfect. Referral traffic from AI tools can be small and inconsistently labelled in analytics, so benefits may show up as brand searches or direct enquiries rather than clean clicks.

Content quality still has to meet legal standards. If you make claims about services or prices, the Advertising Standards Authority's CAP Code expects them to be accurate and not misleading, and that applies equally to pages written to attract AI citations.

Measuring whether you are being mentioned

Because there is no rank to check, you build your own tracking. Keep a fixed list of 15 to 30 questions your customers genuinely ask, run them at regular intervals in both tools, and record whether you are mentioned, how you are described, and which competitors appear. Track the share of answers, not individual wins.

Also watch Google Search Console, since impressions and clicks for question-style queries often move alongside AI visibility. Our walkthrough on tracking brand mentions in AI chatbots covers manual and automated methods, and Illumae includes AI visibility tracking across ChatGPT and Claude if you would rather not do it by hand. Treat any tool's numbers as a trend indicator, not an exact count.

Where to put your effort first

1. Check your robots.txt and security settings. If AI search crawlers are blocked, nothing else you do will be retrieved.
2. Make your business details identical everywhere. Conflicting names, addresses and descriptions give models nothing reliable to repeat.
3. Build a baseline test list. Without a starting percentage, you cannot tell whether later work made any difference.
4. Rewrite your top service pages to lead with the answer. Direct, factual openings are the easiest content for retrieval to lift.
5. Earn independent mentions. Trade bodies, local press and genuine reviews carry more weight than your own claims, so pursue a few good ones over many weak ones.
6. Publish fewer, better pages, then retest monthly. Quality and specificity beat volume, and repeated testing keeps you honest about what is working.