AI search optimization: how to get recommended by ChatGPT, Claude and Google AI

More and more people ask a question to an AI assistant and read one composed answer. ChatGPT, Claude, Gemini, Perplexity, Microsoft Copilot and the AI answers inside Google all work this way.

AI search optimization is everything you do so that those answers mention your business, describe it correctly and link to your site. The assistant picks a handful of names. Your goal is to be one of them.

The good news for a business owner: most of the work overlaps with solid SEO. The new part is a wider view. The assistant reads your website and also everything the web says about you.

You will meet many abbreviations for this field. They describe the same goal from slightly different angles. Pick one term and focus on the work.

The term generative engine optimization comes from a 2023 research paper by a team led from Princeton University. An LLM, or large language model, is the software that writes the answer, such as GPT, Claude or Gemini.

An assistant has two sources of knowledge. The first is training data: text the model learned from up to a fixed date. The second is live search: pages it fetches from the web at the moment you ask.

For questions about local businesses, prices and current products, assistants lean on live search. The assistant turns your question into several search queries, reads the top pages, and writes an answer with links to its sources.

Google calls this technique query fan-out in its Search Central documentation: the system issues multiple related searches across subtopics and combines the results. One question from a customer can therefore trigger ten or more searches.

This explains the first rule of AI search optimization. Pages that rank well in normal search, load as plain text and answer a question clearly are the pages an assistant reads.

Each assistant has its own index, its own crawler and its own taste in sources. This table shows the main places where your customers meet AI answers.

For most small businesses the order of priority is simple: Google first, since it has the largest audience, then ChatGPT, then the others. The steps in this chapter work for all of them.

AI Overviews are the AI-written summaries above the regular Google results. AI Mode is a separate tab where the whole result is a conversation. Google's blog reported in October 2025 that AI Mode had reached more than 200 countries and territories.

Google's Search Central page on AI features is refreshingly plain. It states that there are zero additional requirements to appear in AI Overviews or AI Mode. A page needs to be indexed and eligible to show with a snippet in regular search.

The same page says you can skip new machine-readable files, AI text files and special markup for these features. The foundations from the earlier chapters of this guide are the method.

You keep control through existing tools. The nosnippet and max-snippet settings limit what Google may show from a page. Clicks and impressions from AI Overviews and AI Mode are counted in the Performance report of Search Console, under the Web search type.

On 28 September 2026 we put two questions about 100 Dutch small businesses to Claude, with web search switched on. The sample held 5 businesses in each of 20 industries, spread over 55 towns.

Question one was the discovery question: which business in this industry in this town do you recommend? Question two was the name question: what do you know about this business? These are the headline figures.

Each question was asked once, so the study is a snapshot. AI answers vary per moment and per user. With 5 businesses per industry, differences between industries are an indication.

The most useful finding is the gap between the two questions. When a customer names your business, the assistant searches for you and finds your website. Claude gave at least three correct facts for 88 of the 100 businesses.

When a customer asks for a recommendation, the assistant chooses from every business it meets on the web for that industry and town. Of those 88 well-known businesses, 63 were missing from the recommendation.

Being mentioned also tends to mean being mentioned first: 18 of the 34 recommended businesses held the first position in the answer. On average, each answer named 4.3 other businesses.

We also put the questions to ChatGPT for a subsample of 24 businesses. ChatGPT recommended 13 of them and Claude 9. Both assistants recommended the same business in 8 cases. Visibility is something you earn per assistant.

For the discovery question, Claude consulted 12.1 websites per answer on average, 841 different domains in total. The business's own website appeared among the sources in 43 of the 100 answers.

ChatGPT consulted 30.2 websites per answer on average in the subsample. Reddit appeared among the sources in 17 of those 24 answers.

The names in the table are Dutch. The pattern travels: assistants lean on review platforms, business directories, industry guides and best-of lists for your trade and town. Find the equivalents in your country and claim your place there.

People ask how to rank in ChatGPT as if it had a results page. It works differently: ChatGPT composes an answer and names a few businesses. These seven steps raise the odds that yours is among them.

1. Ask the question yourselfType the question your customer would ask, such as which landscaper in your town it recommends. Write down who appears and which sources it lists.
2. Allow OAI-SearchBotOpenAI's documentation says sites that block this crawler stay out of ChatGPT search answers. Check your robots.txt file.
3. Rank in regular searchThe assistant reads top results. Every chapter of this guide that lifts your rankings also lifts your chance of being read.
4. Claim the sources it consultsCreate complete profiles on the review sites, directories and industry guides that showed up in step one.
5. Collect reviews on those platformsFresh, detailed reviews give the assistant reasons to pick you and words to describe you.
6. State your facts identically everywhereUse the same name, address, phone number and opening hours on your site and every profile.
7. Repeat the test every quarterAnswers change. Track whether you appear, in which position and with which description.

A crawler is a program that fetches your pages. Robots.txt is the text file on your site that tells crawlers what they may fetch. AI companies run separate crawlers for separate jobs, and you can allow or block each one.

Two details from the companies' own documentation. OpenAI says a robots.txt change takes about 24 hours to reach its search systems. Anthropic says its crawlers follow robots.txt, and that blocking by IP address is unreliable.

In our study, all 100 websites allowed the search crawlers of ChatGPT, Claude and Perplexity. Access is rarely the problem for a small business. Check it once, then move on to the steps that make the difference.

An llms.txt file is a plain text file at the root of your site that lists your most important pages with a short description, written for language models. Jeremy Howard of Answer.AI proposed the format in September 2024.

The state of play in 2026: the format is a community proposal. Google's documentation says AI Overviews and AI Mode work from the regular search index and that AI text files are unnecessary for them.

Our own numbers match that picture. In our study 14 of the 100 websites had an llms.txt file. We found zero measurable link between having the file and being recommended.

Our advice: treat llms.txt as optional housekeeping. It takes ten minutes, it does zero harm, and it can help AI coding tools and agents that read documentation. Spend your real effort on content, profiles and reviews.

The difference with robots.txt is worth knowing. Robots.txt gives crawlers rules about access, and the major crawlers follow it. An llms.txt file offers a reading suggestion, and each AI company decides for itself whether to read it.

An assistant lifts short passages from pages and stitches them into an answer. Passages that stand on their own are easiest to lift. The Princeton-led GEO paper found that adding statistics, quotations and cited sources raised visibility in AI answers by up to 40 percent.

1. Answer firstOpen each section with the direct answer in one or two sentences, then give the detail.
2. Use the question as the headingWrite headings the way customers ask, such as how much does a garden design cost.
3. Give exact factsPrices from and to, delivery times, service area, opening hours. Specific facts are quotable, vague claims are skipped.
4. Name your sourcesWhen you cite a number, say where it comes from in the sentence itself.
5. Add tables and step listsComparisons in a table and processes in numbered steps are simple for a model to read and reuse.
6. Show who is speakingAdd the author's name, role and experience. Assistants and Google both favor content from an identifiable expert.

In 29 of 100 answers in our study, the assistant stated a fact that conflicted with the business's own website. Usually it was another address, another town, another phone number or other opening hours.

The cause is old or inconsistent data on the web. An assistant that finds three versions of your phone number has to pick one. You fix this at two levels.

1. On your websiteAdd LocalBusiness or Organization structured data with name, address, phone, opening hours and links to your profiles. Only 11 of 100 websites in our study had this complete.
2. Across the webList every profile that mentions you: Google Business Profile, directories, review sites, social media. Correct each one to match your website exactly.
3. In visible textPut your address, service area and hours as plain text on a contact page and in the footer, so every crawler reads them.

Structured data is code that labels facts for machines. Chapter 7 shows how to add it. Chapter 9 covers Google Business Profile.

An assistant recommends businesses that other sources already recommend. In our study, a review, directory or list site appeared in 81 of 100 answers. Your presence on those sites is the strongest lever you control.

1. Review platforms for your tradeFind the two or three platforms that assistants cite for your industry. Complete your profile and ask every satisfied customer for a review there.
2. Business directoriesClaim your listing on the main national directories. It is dull work that takes an afternoon and lasts for years.
3. Best-of lists and local guidesSearch for best plus your trade plus your town. Contact the publishers of the lists that rank and ask to be considered.
4. Community discussionsChatGPT often cited Reddit in our subsample. Real customer stories on forums and social platforms carry weight. Encourage customers to share their experience in their own words.
5. Local press and partnersA mention on a local news site, a supplier page or a trade association adds another independent source.

Assistants read fast and skip pages that are hard to fetch. The technical list is short and overlaps with chapter 5 of this guide.

1. Text in the HTMLMake sure your content is in the page source. Several AI crawlers skip JavaScript, so text that appears only after scripts run may stay unread.
2. A fast, stable serverA crawler that waits too long moves on to the next source.
3. A sitemap and clean indexingGoogle's AI features need the page to be indexed. Check the Pages report in Search Console.
4. Clear titles and descriptionsName your service and your town. In our study half of the websites had a title or description of the wrong length.
5. Open access for search crawlersCheck that your firewall or security plugin lets the crawlers from the table above through.

One sober note from our study: businesses with a technical score above the median were recommended 35 percent of the time, those below it 33 percent. Technique makes you readable. Reputation makes you recommended.

AI answers vary, so measure with a fixed routine. Write five to ten questions a customer would ask, some with your business name and some leaving it out. Ask them in each assistant and log the result in a spreadsheet.

Add two data sources. Your analytics tool shows visits that arrive from chatgpt.com, perplexity.ai and other assistants as referrals. Search Console includes AI Overviews and AI Mode in your regular Google totals.

Business owners ask whether AI replaces SEO. The evidence points to a merger. Assistants search the web to answer, so ranking pages feed AI answers. Google says the same SEO foundations apply to its AI features.

Treat AI search optimization as the next layer on a solid base. A business with a clear site, consistent facts and a strong review profile is ready for both.

Here is a realistic order of work for a small business. Each block takes a few hours.

• Ask ChatGPT, Claude, Gemini and Google which business in your trade and town they recommend, and log the answers.
• Ask each assistant what it knows about your business and mark every wrong fact.
• List the review sites, directories and guides the assistants cite as sources.
• Check that robots.txt allows OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot and Bingbot.
• Decide whether you allow the training crawlers GPTBot, ClaudeBot and Google-Extended.
• Confirm in Search Console that your key pages are indexed.
• View the source of your home page and check that the main text is in the HTML.
• Add LocalBusiness or Organization structured data with name, address, phone, hours and profile links.
• Make your name, address, phone number and opening hours identical on every profile.
• Claim complete profiles on the two or three review platforms assistants cite for your trade.
• Start a fixed routine for asking customers to review you on those platforms.
• Rewrite your top service pages so each section opens with the direct answer.
• Add exact facts, a table and a short question list to each of those pages.
• Set up a referral view in analytics for visits from AI assistants.
• Repeat the full test every quarter, per assistant.