How AI tools pick what to quote
No company publishes its full method. The official pages do explain the broad steps.
- They read the question and split it. Google says AI Overviews and AI Mode may use "query fan-out", which means many related searches across subtopics. So your page can be used for a small question you never aimed at.
- They need to find your page. Google says a page must be indexed and able to show a snippet to be a supporting link.
- They look for facts they can stand behind. Microsoft says its grounding looks for discrete, supportable facts with clear provenance. Provenance means where a fact came from.
- They may decline to answer. Microsoft says a system should abstain when the evidence is missing, out of date or in conflict.
- They show sources. Bing's AI Performance report lists the pages cited and the phrases the AI used to find them.
Our reading: plain, specific, current and consistent facts give a tool something safe to use. Vague claims and old figures do not.
Step 1: Let the right bots in
If a tool cannot read your page, it cannot quote it. Check three places.
- Your robots.txt file. This is the plain text file at the root of your site that tells crawlers what they may read.
- Your host, firewall or content delivery network (CDN). Google lists CDN and hosting blocks next to robots.txt as things to check. OpenAI and Perplexity both ask you to allow their published address lists.
- Your page settings. A noindex tag or a very tight snippet limit keeps a page out of Google's AI features. Google says more restrictive snippet settings limit how your content is featured.
OpenAI says a robots.txt change can take about 24 hours to take effect. It also says its search bot and its training bot are separate settings. You can allow one and block the other.
Step 2: Write the key facts in plain text
Google says important content should be available in text form. Bing says clear headings, tables and FAQ sections make content easier for AI to use.
- State who you are, what you sell, where you work, who it is for and how to reach you. Put these on pages a shopper would visit.
- Use the question as a heading and answer it in the first lines below. Add the detail after.
- Say prices in kind if you cannot give figures: what drives the cost, and what a quote includes.
- Add dates to anything that changes: rules, offers, hours, versions.
- Keep one name for each product and service. Do not call the same thing three names.
Say a coaching centre has a question that parents ask on calls: "When does the next batch start?" The page puts that question as a heading. The first line gives the date. That is easy to lift, and easy to check. The example is made up.
Step 3: Keep facts true and the same everywhere
Google tells site owners to keep their Business Profile and Merchant Center details up to date. Bing says to register with Bing Places so address, hours and contact details stay current.
- Check your name, address, phone number and hours on your site, your Google Business Profile, your Bing Places listing and your main directories.
- Remove old pages or mark them with a date. Bing says to keep content fresh and to reduce ambiguity across text, images and video.
- If you change a page often, tell search engines. Bing points to IndexNow for this.
Step 4: Show proof
Google's own questions for useful content ask for original information, first-hand experience, clear sourcing and information about the author or the site. Bing's advice is similar: back claims with examples, data and cited sources.
- Name the person or team who wrote the page, and what they know.
- Link to the official source for any rule or number.
- Show real work: photos, process, results you can prove.
- Do not copy and reword other pages. Google lists that as a warning sign.
Reviews and mentions on other sites are a common talking point. We found no official page from Google, OpenAI, Perplexity, Anthropic or Microsoft saying they make an AI tool cite you. Reviews still matter to the customer, so ask real customers and answer them.
Step 5: Use structured data only to match the page
Structured data, or schema, is code that labels facts on a page for machines, such as an address or a product. Google says it needs no special schema for its AI features. It also says any markup must match the text a visitor can see. So label what is on the page. Do not add claims that are not.
What about llms.txt?
llms.txt is a proposal for a plain file at the root of a site. It gives a short guide and links to pages written for AI tools to read. Its own site says it is used most for software documentation. Read our llms.txt glossary page for the format.
Google's documentation says you do not need new machine readable files, AI text files or markup to appear in its AI features. We searched for an official statement from OpenAI, Anthropic or Perplexity that llms.txt changes who they cite, and found none.
So treat it as optional and low priority. If a developer can add it in an hour, fine. Do not pay a fee for it as a way to win citations.
What not to believe
- "We guarantee you will be cited." Google says indexing and serving are not guaranteed. No agency controls an AI tool.
- "A secret file or tag opens the door to AI search." Google says no special files, schema or markup are needed for its features.
- "Research shows GEO lifts visibility by 40%." The GEO paper reported gains of up to 40% in its own tests. It also said results differ by field. Those were test queries, not your business.
- "Our dashboard shows your AI rank." Answers change with the question, the person and the day. Look for patterns, not a rank.
- "Block all AI bots to be safe." That can remove you from the tools that send you customers. Choose bot by bot.
How to check where you appear
- Write down 10 questions your customers really ask. Take them from sales calls and WhatsApp chats.
- Ask each one in Google (AI Mode), ChatGPT search, Perplexity, Copilot and Claude with search. Write down whether you were mentioned, linked or missing, and whether the facts were right.
- Repeat on the same day each month. Keep the date, tool and exact words. Answers vary, so judge the pattern.
- In Google Search Console, open the Performance report. Google says AI features are counted in the "Web" search type, so they are not shown separately.
- In Bing Webmaster Tools, open AI Performance. It shows total citations, pages cited and the grounding phrases, from a sample of activity.
- Count leads, not mentions. Ask each new enquiry how they found you, and log it in your own lead record.
A mention that brings no enquiry is a nice screenshot, not a result.
Mistakes we see
- Paying for a special file or tag before the basic pages are clear.
- Blocking a crawler by accident through a firewall or an old robots.txt rule.
- Letting prices, hours and offers go out of date on the site.
- Writing in a style that sounds the same as every other page.
- Judging the work by one screenshot of one answer.
- Forgetting the customers who ask in Hindi or another Indian language.
When to ask for help
Ask for help when the basics are done and you want an honest check of where you stand, or when your site is large. We look at crawl access, page facts, proof and tracking together. See our AI SEO service, and the pages on ChatGPT optimisation and Google AI Overviews. For more on Google's side, read how to appear in Google AI Overviews.
