Can You Control Whether AI Systems Cite or Recommend Your Business?
You can control whether AI systems exclude your content, but not whether they cite or recommend your business. Site owners have real switches for keeping content out. There are robots rules for each crawler and snippet controls on the page. Search Console has a setting for Google's generative AI features, and Google-Extended covers Gemini training and grounding. Getting cited or recommended can only be influenced, through your pages, profiles, reviews and third-party coverage. Even then, answers vary between runs as models, sources and competitors change. Tracking tells you where you stand and where to act, but it does not guarantee that any fix will work, or when. Anyone who promises organic placement in AI answers is promising something no one controls.
What can you control directly?
| Control | What it does |
|---|---|
| robots.txt rules per crawler | Allow or block each AI company's training crawler and search crawler separately. User-triggered fetchers are less reliable: Perplexity says Perplexity-User generally ignores robots.txt, and OpenAI says robots.txt may not apply to ChatGPT-User, so block those at the server or firewall if you must |
nosnippet, data-nosnippet, max-snippet |
Limit what Google shows from a page, including in AI features; Bing added data-nosnippet support in 2025 |
| Search Console's generative AI setting | Include or exclude your site from Google's AI Overviews, AI Mode and generative features in Discover |
| Google-Extended | Opt out of your content being used for Gemini training and grounding in the Gemini app, without affecting Search |
| Your own pages, facts and markup | What your site says, how clearly, and whether crawlers can read it |
| Your profiles | Business Profile, directory listings and company profiles you own |
These are real, reliable controls, but they mostly work in one direction. You can stop an assistant from reading or showing your content. You cannot make it recommend you.
What can you influence?
- Third-party coverage: reviews, roundups, comparison articles, press and community discussion.
- Accuracy of sources: asking publishers and directories to correct errors.
- Reports to platforms: feedback and reporting routes for false or harmful answers.
- Relevance: pages that answer the specific questions buyers ask, with facts others lack.
- Consistency: the same facts across every source assistants read.
Influence works on probabilities. A clear page and three good third-party mentions make it more likely you are recommended for a question, not certain. How assistants choose which brands to recommend is covered in how AI systems decide which brands to recommend.
What is out of your reach?
- What the model already learned. Answers without a web search draw on training data you cannot edit.
- How answers are composed. Which sources are combined, and how they are summarised.
- Randomness. The same question can produce different answers. Ahrefs found AI Overviews had a 70% chance of changing between one observation and the next.
- Platform changes. seoClarity found citation volumes in ChatGPT answers fell by 86% to 94% over two months in 2026, then recovered, for every brand at once.
- Competitors, and the sources that favour them.
- How the question is asked. Buyer phrasing can change recommendations more than anything on your site, as covered in whether buyer phrasing changes AI recommendations more than the platform you use.
Does AI visibility tracking guarantee you can fix what models recommend?
No. Tracking measures where you appear, how you are described and which sources are cited. That is essential for knowing where to act, but it is a diagnosis, not a cure. Whether a fix works depends on the cause:
| Cause found by tracking | Can you fix it? | Typical time |
|---|---|---|
| Your page is blocked, unreadable or wrong | Yes | Weeks |
| A directory or profile you control is wrong | Yes | Weeks |
| A third-party article is wrong | Often, if the publisher corrects it | Weeks to months |
| You are missing from cited roundups | Sometimes, on merit | Months |
| Competitors are simply stronger for the question | Only by becoming a better option | Months or longer |
| The model's training knowledge is outdated | Not directly | Until the model is updated |
A good tracking programme separates these causes, so effort goes where it can succeed. A report that shows only a visibility score, without causes, cannot tell you what to fix.
How should you set expectations?
- Talk about rates, not positions: "mentioned in 30% of answers to these 40 questions", measured over repeated runs.
- Separate what you control from what you influence in plans and reports.
- Set timescales by cause: fast for your own sources, slower for third-party and model knowledge.
- Expect variation, and judge trends over months rather than a single report, especially after platform changes that affect every brand.
- Be wary of guarantees, from vendors or from anyone else.
Which controls reach which assistants?
Each company names its crawlers separately, so exclusion has to be set per assistant:
- Google Search, AI Overviews and AI Mode use Googlebot. Blocking Googlebot stops Google crawling your pages, so they lose their snippets and nearly all Search visibility. To stay in Search but out of the AI features, use Search Console's generative AI setting or snippet controls.
- The Gemini app grounding and Gemini training follow Google-Extended.
- ChatGPT uses OAI-SearchBot for search and GPTBot for training, with ChatGPT-User for fetches a user triggers.
- Claude uses Claude-SearchBot for search, ClaudeBot for training and Claude-User for user fetches.
- Perplexity uses PerplexityBot for its index and Perplexity-User for user fetches.
- Microsoft Copilot draws on Bing, so Bingbot rules and Bing's snippet controls apply. Microsoft's webmaster guidelines now describe how each directive affects Copilot answers.
Blocking a training crawler while allowing a search crawler is a common middle path. Your content stays out of future training but can still be cited in live answers.
When is exclusion the right choice?
Sometimes you want less visibility, not more. Examples include content you license, pages behind a paywall, internal tools, or information that should not be summarised out of context. Use the controls above deliberately. Blocking an AI company's search crawler also removes your pages as a source for that assistant's live answers, so consider the trade-off for each crawler, rather than blocking everything at once.
What might an illustrative case look like?
Consider an illustrative furniture retailer whose leadership asked the marketing team to "get us into ChatGPT for sofas" by the end of the quarter. The team's tracking showed the brand in 12% of answers to 40 sofa questions. The main cited sources were two review sites and a comparison article that did not include it. The team set out what it controlled: product pages, delivery and returns facts, and its markup. What it could influence were the review sites, the comparison article's author and customer reviews. It explained what it could not control, and reported a mention rate over repeated runs. By the end of the quarter the rate was 19%, and leadership judged the work on that trend rather than a single screenshot.
What do people ask about controlling AI answers?
Can paying an AI company get us recommended?
Paying does not change the assistant's own organic picks: OpenAI and Microsoft say ads do not influence ChatGPT or Copilot answers, and their ads are shown separately. Google's AI Mode is different: its Highlighted Answers format lets ads appear inside recommendation lists, labelled "Sponsored". That is a paid placement, not an organic recommendation.
Can we remove ourselves from AI answers entirely?
Largely, for assistants that respect your crawler and snippet controls. Model knowledge from earlier training may persist for a time.
Is it worth trying if we cannot control the outcome?
Yes. Most of what improves your chances, such as accurate pages, good reviews and credible coverage, also helps with customers directly.
Published by The Kovmere Team · updated Sep 29, 2026 · More resources