Learn
The sources behind our GEO practices.
GEO advice should rest on evidence. These are the studies, platform guides, standards and rules our methods follow.
In short
24 sources in 5 groups: research on AI answers, what Google and Microsoft say about AI search, how each AI company's crawlers reach your site, structured data, and the rules on reviews and endorsements. Each one says what it is, what it says in our words, and what we take from it. Links checked October 3, 2026.
1 of 5
Research on AI answers
Independent studies of how generative search chooses and cites sources, and how people use it.
- GEO: Generative Engine Optimization
The paper that named the field. In its experiments, changes to how content was written raised a source's visibility inside generated answers by up to 40%, and what worked varied by subject.
What we take from it: How a page is written matters, and what works differs by subject — so we measure your market instead of applying one template.
- Generative Engine Optimization: How to Dominate AI Search
AI search engines systematically favor earned media — third-party, authoritative sources — over brand-owned and social content, unlike Google's more balanced mix, and differ from one another in which domains they use, how fresh their sources are and how sensitive they are to phrasing.
What we take from it: Why independent evidence (reviews, articles, press, listings) matters as much as your own pages, and why we measure each assistant separately.
- Google users are less likely to click on links when an AI summary appears in the results
In March 2025 about 18% of the Google searches studied produced an AI summary. People clicked a traditional result in 8% of visits with a summary versus 15% without, and clicked a link inside the summary in 1% of visits.
What we take from it: Being named and described correctly inside the answer matters, not only ranking below it.
2 of 5
What search engines say about AI answers
Google's and Microsoft's own guidance for appearing in AI Overviews, AI Mode, Copilot and Bing's AI answers.
- Google's guide to optimizing for generative AI features on Google Search
Search fundamentals still apply: crawlable, indexable pages; unique, non-commodity content with real expertise; clear page structure; and accurate Business Profile and Merchant Center details. No special AI text files, markup or Markdown are needed, and Google Search ignores llms.txt.
What we take from it: Fundamentals first. We treat llms.txt as an optional extra, never as a substitute for clear pages.
- AI features and your website
There are no additional technical requirements for AI Overviews or AI Mode: a page must be indexed and eligible for a snippet. Site owners can limit what is shown with nosnippet, data-nosnippet, max-snippet or noindex.
What we take from it: Every audit starts by checking that your pages can be crawled, indexed and quoted.
- Top ways to ensure your content performs well in Google's AI experiences on Search
Google's own checklist for AI Overviews and AI Mode: unique, valuable content made for people; a good page experience; pages Google's systems can reach; preview controls; structured data that matches what the page shows; and images and video alongside text.
What we take from it: The checklist our website fixes follow, page by page.
- Creating helpful, reliable, people-first content
Content should be made primarily for people — original, complete and clearly sourced — rather than to manipulate rankings.
What we take from it: We write for your buyers first; anything written only for machines is left out.
- Optimizing Your Content for Inclusion in AI Search Answers
AI systems break pages into small pieces: clear headings, schema, question-and-answer sections, lists and tables help; so does precise language anchored in measurable facts, and concise answers that make sense on their own.
What we take from it: The format of our pages and our clients' pages: the answer first, a key-facts table, one question per heading.
- Introducing AI Performance in Bing Webmaster Tools (public preview)
Shows when a site is cited as a source in Microsoft Copilot and Bing's AI summaries, including total citations and page-level citation activity.
What we take from it: One more signal we read alongside our own measurements.
3 of 5
How AI assistants reach your website
Each AI company documents its own crawlers. Search crawlers decide whether you can be found and cited; training crawlers are a separate choice.
- Overview of OpenAI crawlers
OAI-SearchBot surfaces websites in ChatGPT search and follows robots.txt (changes take about a day to apply); GPTBot collects content for model training; ChatGPT-User fetches pages for actions a user asks for.
What we take from it: Allow OAI-SearchBot if you want to be found in ChatGPT search; whether to allow training is a separate decision.
- Does Anthropic crawl data from the web, and how can site owners block the crawler?
ClaudeBot collects training data. Claude-SearchBot and Claude-User fetch pages for Claude's search and for users' requests, and blocking them can reduce a site's visibility in Claude's answers.
What we take from it: Allow Claude-SearchBot and Claude-User to be found and cited in Claude.
- Perplexity Crawlers
PerplexityBot surfaces and links websites in Perplexity's search results and is not used to train foundation models; Perplexity-User visits pages to answer users' questions.
What we take from it: Allow PerplexityBot to be eligible for Perplexity's citations.
- About Applebot
Applebot powers search in Siri, Spotlight and Safari. Applebot-Extended does not crawl; it controls whether pages Applebot collected can train Apple's foundation models.
What we take from it: Another case where search visibility and AI training are separate settings.
- RFC 9309: Robots Exclusion Protocol
The standard for robots.txt: user-agent lines name crawlers, allow and disallow rules set which paths they may visit — a courtesy for crawlers, not access control.
What we take from it: How we write robots.txt rules that each AI crawler reads the same way. Ours names every major one.
4 of 5
Structured data and machine-readable files
Markup and files that help machines read a page's facts — useful only when they match what the page says.
- Intro to how structured data markup works
Structured data helps Google understand a page. JSON-LD is the recommended format; include accurate required and recommended properties rather than every possible field, and validate it.
What we take from it: We add JSON-LD that repeats what the page visibly says — never more.
- Organization structured data
Organization markup describes a business's administrative details and logo; sameAs links its profiles on other sites so it can be told apart from others with similar names.
What we take from it: Your name, logo and profiles stated once, the same way everywhere.
- Review snippet structured data
Reviews a business selects and hosts about itself are treated differently from independent reviews.
What we take from it: We never mark up your own testimonials as if they were independent reviews.
- The /llms.txt file
Proposes a Markdown file at /llms.txt that gives language models a short, curated map of a website.
What we take from it: We publish one as an example; it is optional, and Google says its search ignores it.
- IndexNow
Lets a website tell search engines straight away that pages were added, changed or deleted; used by Microsoft Bing, Naver, Seznam.cz, Yandex and Yep.
What we take from it: A quick way to get changed pages re-read after a fix goes live.
5 of 5
Reviews, endorsements and links: the rules
The rules our reputation work follows. Genuine evidence is what AI should repeat — and the only kind that lasts.
- Trade Regulation Rule on the Use of Consumer Reviews and Testimonials (16 CFR Part 465)
In effect since October 21, 2024: bans fake reviews (including AI-generated ones), paying for reviews that must be positive or negative, undisclosed insider reviews and suppressing negative reviews.
What we take from it: We never write, buy or filter reviews.
- FTC's Endorsement Guides: What People Are Asking
Endorsements must reflect honest opinions, and any connection consumers wouldn't expect — payment, free products, commissions — must be disclosed clearly.
What we take from it: Mentions we help earn are genuine, and any connection is disclosed.
- Maps User Generated Content Policy — Prohibited and restricted content
Businesses may not offer incentives for reviews, discourage negative reviews or ask only happy customers.
What we take from it: We ask every customer the same way, never only the happy ones.
- Spam policies for Google web search
Buying or selling links for ranking purposes — including paying for posts that contain links — breaks Google's policies.
What we take from it: Coverage is earned, never bought.
- Tips to improve your local ranking on Google
Local results are mainly based on relevance, distance and popularity; complete, accurate information and more reviews and positive ratings can help.
What we take from it: Why accurate listings and genuine reviews are part of GEO for local businesses.
AI models
Which AI models we monitor
The models we test are documented on each provider's own pages; we list them with links on Deep monitoring. How we use all of this: How we measure and What is GEO?
Next step
See how AI assistants describe your business today.
Create your free account in a minute — it starts nothing. When you're ready, choose a plan and pay at checkout; your results appear in your portal.