How to show up in ChatGPT: a guide for businesses

First, how showing up in ChatGPT works

When someone asks ChatGPT “which online accountant would you recommend for a freelancer?”, one of two things can happen. If the model answers from what it learned in training, your presence depends on what was on the web when it was trained. If the model searches the web before answering (ChatGPT with search, Perplexity, Gemini, AI Overviews), your presence depends on whether your pages, and the pages that talk about you, are crawlable, clear and relevant to that question, right now.

The second route is the one you can work on. That is why almost everything below looks like running a good website: not because GEO is SEO with another name, but because AI with search retrieves documents before it generates an answer, and it can only cite what it retrieves.

1. Let the right crawlers in

OpenAI uses different crawlers for different purposes. OAI-SearchBot indexes content for ChatGPT search; GPTBot is used to train models. Any public website can appear in ChatGPT search, but if you block OAI-SearchBot in robots.txt it will not be able to discover or cite you. You can allow one and block the other: they are separate decisions.

Check your robots.txt and your server or CDN headers. Many websites block “all AI bots” with a blanket rule, search included. Google, for its part, says no special file or markup is needed to appear in its AI features: pages are crawled and indexed as they are for normal search.

2. Check that you exist for search engines

Before you write a single new line, confirm three things:

• Indexing. Your main pages show up on Google when you search site:yourdomain.com. If they do not, you have a technical problem before a GEO problem.
• Crawling. Search Console shows no coverage errors on the pages that matter, and you do not depend on JavaScript to display the essential content.
• One clear sentence. The home page says, in visible text, what you do, for whom and where. “Comprehensive solutions to drive your business forward” says nothing. “Online accounting for freelancers in Madrid, with VAT and income tax returns filed for you” does.

3. Publish the pages that answer what people ask

When we look at the pages cited in AI answers where other brands appear, the same types keep coming up: comparisons, pricing pages, FAQs, “how to choose” guides, directories and reviews. Rarely a home page.

Start with these four:

1. Pricing. “How much does it cost?” is the question where businesses lose out most. If your answer is “request a quote”, the AI will most likely cite whoever does state a price. Publish ranges, what each one includes and what it does not.
2. Buying FAQs. Not the logistics ones (“how do I change my password?”) but the ones people ask before deciding: differences, requirements, timelines, guarantees, who it is not for.
3. An honest comparison. Your product against the alternatives people already compare it with. Verifiable facts, no adjectives. If your competitor is better at something, say so: the page’s credibility depends on it.
4. A complete service or product page. What it is, who it is for, how it works, how long it takes, what the client needs to provide, what it does not include.

Each page should answer in its first lines. Our reading: if the answer is in paragraph seven, it is less likely to be cited.

4. Be the same business everywhere

Models build an idea of your entity from many sources. If your website says “Smile Dental Clinic”, your Google Business Profile says “Smile Dental” and a directory says “Smile Clinic Valencia”, you have just split your brand into three. The same goes for addresses, phone numbers, services and descriptions.

Run a consistency audit: exact name, one-sentence description, address, category and services identical on your website, Google Business Profile, LinkedIn, industry directories and official profiles. Add Organization markup to your website with true data (name, URL, logo, address) and, if your website shows the people behind the business, Person markup. Do not invent ratings, awards or identifiers.

5. Appear in the sources AI already cites

Measure first (the twenty-minute method will do) and note which domains keep coming up in the answers for your category. They may be comparison sites, specialist media, communities or directories. There you have three legitimate options:

• Directories and comparison sites: check that your listing exists, is correct and is complete. Ask for a correction if there is an error.
• Industry media and newsletters: offer data, a guide or real experience, not a link. If there is a commercial arrangement, it is labelled as such.
• Communities: take part with your name and your company visible, answering real questions. Never with fake profiles.

What does not work, and exposes you as well: bought reviews, coordinated votes, undisclosed paid links and accounts pretending to be customers. Apart from the legal risk, a platform can detect it and an AI can end up repeating it.

6. Look after what other websites say about you

If AI describes you wrongly, it is often copying a source. Find it: an old listing, an outdated article, an ambiguous page of your own. Fix what you control first; then ask third parties for a correction, with the correct source linked. It is slow, and it works better than arguing with the model.

7. Measure before, change, measure after

None of the above helps if you cannot see the effect. Save twelve buying questions, two platforms and the full answers with their dates. Make the changes. Repeat the same questions two or three weeks later. Compare answer with answer, not impression with impression.

And accept what the measurement does not prove: a difference between before and after coincides with your changes, but it does not certify them as the cause. With three or four rounds you start to tell a trend from noise.

What you do not need

• An llms.txt file. Google says no special file is needed for its AI features. You can experiment, but it is not a requirement.
• Rewriting the website “for AI”. Write for the person asking. It is the same text.
• Repeating the keyword a hundred times. Models with search retrieve by relevance and clarity, not by density.
• Schema tricks. Structured data describes what is already visible. It does not turn an empty page into a source.

The 20-point checklist

Access

1. robots.txt allows OAI-SearchBot and the search crawlers you want.
2. No CDN or firewall rule blocks those crawlers.
3. Key pages return 200 and do not need JavaScript for the main content.
4. An up-to-date sitemap, submitted in Search Console and Bing Webmaster Tools.

Clarity

5. The home page says what you do, for whom and where on the first screen.
6. Each service or product has its own page, with a title that names it.
7. There is a pricing page with real ranges.
8. There is an FAQ with buying questions, not only support ones.
9. There is at least one honest comparison with alternatives.
10. Each page answers in its first lines.

Entity

11. Name, description, address and category identical on your website, Google Business Profile, LinkedIn and directories.
12. Organization markup with true data.
13. An “about us” page with real people, names and roles.
14. A visible author and date on articles.

Sources

15. You know which domains AI cites in your category (you have measured them).
16. Your listing exists and is correct on the relevant directories and comparison sites.
17. You have offered something useful to an industry publication or newsletter in the last three months.
18. You take part in communities under a real identity.

Measurement

19. You have twelve unbranded questions saved, with dates and full answers.
20. You have a date set to repeat them.