Generative engine optimization (GEO) is the work of getting a brand found, cited and named in the answers that ChatGPT, Gemini, Perplexity, Microsoft Copilot and Google's AI Overviews and AI Mode give. What to do about it fits in one sentence: make sure the engines can index and crawl your pages, publish the clearest citable answer to the questions your buyers ask, earn mentions on the sites the engines already cite, and measure with the free first-party reports from Google and Microsoft.

Most of that is SEO, and Google says so in its own guidance. What changed is the unit of competition. A results page competes for a click. An AI answer pulls a few passages from a handful of sources and often sends no click at all, so the goal moves from ranking a page to being the source a passage comes from. This guide is for marketing leaders at B2B and B2B SaaS companies deciding what to change now that AI answers absorb a growing share of clicks.

Key takeaways:

  • Google's AI optimization guide (May 2026) says optimizing for AI Overviews and AI Mode is still SEO, with no special schema.org markup or llms.txt file needed.
  • A page has to clear three gates before its content matters: it must be in the engine's index, crawlable by the engine's retrieval bot and eligible to show a snippet.
  • AI summaries cut clicks. Pew found people clicked a traditional result on 8% of visits to a Google results page with an AI summary, against 15% without one (March 2025 data).
  • In Ahrefs' studies of 75,000 brands, branded web mentions correlated with AI visibility far more strongly than backlinks did, though Ahrefs stresses that correlation isn't causation.
  • Measure AI visibility as a share across many runs of the same prompts, never as a rank, starting with Search Console's Generative AI report and Bing's AI Performance report.

What is generative engine optimization, and how is it different from SEO, AEO and "AI SEO"?

Generative engine optimization is the practice of improving how often, and how accurately, AI answer engines cite and mention a brand. It overlaps almost entirely with SEO. Answer engine optimization (AEO) is a near-synonym, and "AI SEO" is ambiguous: it can mean optimizing for AI answers or using AI tools to do SEO work.

The term comes from "GEO: Generative Engine Optimization" (Aggarwal et al.), a 2023 paper by three researchers at Princeton University, one at IIT Delhi and two working independently, published at KDD 2024. It defined a generative engine as a system that retrieves sources and writes one answer with inline citations, which describes AI Overviews, AI Mode, ChatGPT search, Perplexity and Copilot today.

TermWhat it means
Generative engine optimization (GEO)Improving a brand's visibility in AI answers that cite sources
Answer engine optimization (AEO)The same goal under an older name, rooted in featured snippets and voice answers
AI search optimizationAn umbrella phrase for GEO and AEO
LLM optimization (LLMO)Shaping what models say, including what they recall from training data
AI SEOEither GEO, or using AI tools to do SEO tasks, so check which one is meant

For a shorter primer, see GEO basics. If you came for the second meaning of AI SEO, using AI tools to produce the work, AI content marketing covers that ground.

How is GEO different from SEO?

GEO and SEO share one foundation: indexing, crawling and helpful content. The difference is what wins. SEO competes for a ranked position and the click that follows. GEO competes to be one of the few sources an answer draws from, which gives quotable passages and off-site mentions more weight than they carry in classic rankings.

Traditional SEOGenerative engine optimization
GoalRank a page and earn the clickBe a cited source, or the brand named, inside the answer
Unit that competesThe pageThe passage, plus what other sites say about the brand
Where it's wonYour site and the links pointing to itYour site, plus the third-party pages engines retrieve
Main metricsRankings, clicks, impressionsCitations, mention rate across repeated prompts, AI referrals
First-party reportsSearch Console and Bing Webmaster Tools performanceSearch Console's Generative AI report and Bing's AI Performance
Best forQueries where people still click: navigational, transactional, deep researchInformational and comparison questions where the answer is the destination

Google's position is blunt: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO." I agree on the mechanics. I disagree with anyone who reads that line as permission to change nothing. The work is SEO, but the scoreboard changed, and a page that ranks third can still lose the answer to a forum thread and a competitor's comparison page.

Is answer engine optimization the same as GEO?

In practice, yes. Answer engine optimization started with featured snippets and voice assistants and now describes the same goal as GEO: being the answer an engine shows. Google's guide treats AEO and GEO together and files both under SEO. The answer engine optimization playbook covers the AEO framing and the ANSWER framework in more depth.

How do AI answer engines choose which sources to cite?

AI answer engines retrieve pages from a search index, pick the passages that best support each part of the answer and cite them. Google names two techniques: retrieval-augmented generation, which grounds the answer in pages from its index, and query fan-out, which issues related sub-queries. A page enters that pool only after clearing three gates.

Fan-out matters for B2B buyers because their questions are compound. A question like "best SOC 2 compliance software for a 50-person startup" can trigger separate retrievals for pricing, integrations and alternatives, and each sub-query is a separate chance to be cited.

  1. Indexed. Google says a page must be indexed and eligible to be shown in Google Search with a snippet to appear in its generative AI features. Microsoft says Copilot is powered by Bing's search index, so Bing indexing matters as much for Microsoft's answers.
  2. Crawlable by the retrieval bot. Each engine runs its own crawler, and one robots.txt rule or CDN bot-blocking setting can remove a site from an engine without anyone noticing. OpenAI says sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers.
  3. Extractable. Google's nosnippet and data-nosnippet controls limit what it may quote, and the Search generative AI control in Search Console has to be set to include, which is the default. Text hidden in tabs, PDFs or images is harder for any system to lift.

Each engine documents its index and crawler differently. This table comes from the vendors' own documentation:

EngineIndex it relies onCrawler or control tokenWhat it appears to reward
Google AI Overviews and AI ModeGoogle Search indexGooglebot, plus the Search generative AI controlIndexed, snippet-eligible, "non-commodity" content, per Google's guide
Gemini appGoogle Search index, for groundingGoogle-Extended, for training and groundingNo separate guidance
Microsoft Copilot and Bing AI answersBing indexBingbotClear titles, question headings, lists and tables, per Microsoft's guidance
ChatGPT searchNot fully described in OpenAI's crawler docsOAI-SearchBot for search, GPTBot for training, ChatGPT-User for user requestsNo published content guidance
PerplexityIts own index, built by PerplexityBotPerplexityBot for search, Perplexity-User for user requestsNo published content guidance
ClaudeNot named in Anthropic's crawler docsClaude-SearchBot for search, ClaudeBot for training, Claude-User for user requestsNo published content guidance

The last column is thin for three engines on purpose. OpenAI, Perplexity and Anthropic publish crawler rules, not ranking guidance, and anyone claiming to know their weighting is guessing. Content quality only matters after the gates, so I'd audit indexing and crawler access before rewriting a single page. The same order applies to how to rank in AI Overviews.

AI search platform ecosystem with floating orbs representing Google, ChatGPT, Perplexity

How much is AI search changing clicks and traffic?

AI answers are cutting clicks, not searches. Pew found a click on a traditional result on 8% of visits to a Google results page with an AI summary, against 15% without one, and Ahrefs measured a position-one click-through rate 58% lower with an AI Overview on December 2025 data. Meanwhile, Google reported record query volume in 2026.

The reach is large. AI Mode opened to all U.S. users on May 20, 2025, and at I/O on May 19, 2026, Google said AI Mode had passed one billion monthly users and that queries had hit an all-time high the previous quarter. OpenAI launched SearchGPT as a prototype on July 25, 2024 and folded it into ChatGPT as ChatGPT search on October 31, 2024.

How often AI summaries appear depends on who is counting. Pew measured them on 18% of Google searches in March 2025 (12,593 of 68,879 searches by 900 U.S. adults). McKinsey's October 2025 trend analysis put AI Overviews at about 50% of searches and expected more than 75% by 2028. One is a measurement of a panel's real searches and the other is a projection, so quote both with their dates.

  • Pew (July 22, 2025): people clicked a link inside the AI summary on 1% of visits, and ended their browsing session on 26% of results pages with a summary, against 16% without one.
  • Ahrefs: position-one click-through rate was 34.5% lower with an AI Overview in its April 2025 study and 58% lower on December 2025 data, published February 4, 2026.
  • SparkToro (June 2026): 68.01% of U.S. Google searches ended without a click from January to April 2026, on Similarweb panel data. The 2024 figure, 60.45%, came from Datos, and SparkToro cautions that the two panels "aren't the same users or devices." Only 0.34% of searches moved into AI Mode.

The traffic collapse some predicted hasn't arrived. On February 19, 2024, Gartner predicted that search engine volume would drop 25% by 2026. Datos' Q1 2026 report, as covered by PPC Land, had AI tools at 1.65% of U.S. desktop activity, still under 2%. StatCounter had Google at 91.05% of worldwide search in October 2026, with Bing at 4.8%. Adoption keeps growing anyway: Menlo Ventures (September 16, 2026) found 64% of U.S. adults have used AI, up from 61% a year earlier.

For a B2B marketer, the practical change is fewer clicks on informational queries. Top-of-funnel content that used to bring visits now has to win the citation and the brand mention, because the click may never come. The AI Mode zero-click news post and the guide to AI in search engines cover the product changes behind these numbers.

Transformation from traditional SEO traffic to AI citations with melting website structures

Do you need schema, llms.txt or special markup to appear in AI answers?

No. Google's guide says structured data isn't required for its generative AI features, that there's no special schema.org markup to add and that Google Search doesn't use llms.txt files. Schema still helps machines understand entities and qualify for rich results, so treat it as hygiene, not as the lever that gets a page cited.

Google's wording: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add." Google still recommends structured data for rich results. Microsoft's October 2025 guidance says schema "helps search engines and AI systems understand your content." Both are true at once. Schema labels what a page is, and neither company says it decides which page gets cited.

  • llms.txt: Google says maintaining one "will neither harm nor help your site's visibility or rankings in Google Search, as Google Search ignores them." Other services may read the file, and the llms.txt explainer covers when it's worth keeping.
  • FAQ markup: Google stopped showing FAQ rich results on May 7, 2026. A visible FAQ section still helps readers and gives engines clean question and answer pairs, so keep FAQPage markup as hygiene, not as a ranking play.
  • Chunking: Google says there's no requirement to break content into tiny pieces for its AI systems, while Microsoft says Copilot parses pages into smaller pieces before assembling answers. Sections that each answer one question serve both, and they serve the reader first.

What makes a page more likely to be cited?

A page gets cited when one of its passages answers a specific question plainly, carries evidence a model can attribute and still makes sense when lifted out of the page. Google calls this non-commodity content with a unique point of view, Microsoft calls it snippable content, and the Princeton GEO tests found citations, quotes and statistics raised visibility most.

Google's guide draws the line with an example: commodity content, "something like '7 Tips for First-Time Homebuyers'", restates common knowledge, while non-commodity content offers "unique expert or experienced takes." The B2B version is the gap between yet another "What is product-led growth?" page and a page with your own benchmark of trial-to-paid conversion by segment.

  • Open each section with a 40 to 60 word answer to its heading, then give the evidence.
  • Write headings as the questions buyers ask. Microsoft's example replaces "Learn More" with "What Makes This Dishwasher Quieter Than Most Models?"
  • Make each paragraph stand on its own, and name the subject at the start of a section instead of opening with "it".
  • Put a date and a linked source on every statistic.
  • Quote named people and give their roles.
  • Use tables for comparisons and numbered lists for steps.
  • Keep key facts in HTML text. Microsoft warns that AI systems may not render content hidden in tabs or expandable menus, that PDFs often lack the headings and metadata HTML provides, and that pulling text out of images "often reduces accuracy."

Microsoft's Krishna Madhavan described the shift this way: "In AI search, ranking still happens, but it's less about ordering entire pages and more about which pieces of content earn a place in the final answer."

The Princeton paper put numbers on these writing choices. On its GEO-bench test set, adding citations, quotations or statistics lifted a source's visibility by 30 to 40% on the paper's position-adjusted word count measure and by 15 to 30% on its subjective impression score. Keyword stuffing scored below the unoptimized baseline on position-adjusted word count, 17.7 against 19.3, and only slightly above it on subjective impression, at 20.2. Lower-ranked pages gained the most: citing sources lifted the fifth-ranked source by 115.1% while the top-ranked source lost 30.3%. The tests used a GPT-3.5 engine over the top five Google results, with a confirming check on Perplexity, so read them as direction rather than forecast.

For engine-specific tactics, see how to optimize content for ChatGPT citations and the CITE framework in AI search optimization across engines.

Why do brand mentions matter more than backlinks in AI answers?

Because AI answers draw mostly on pages a brand doesn't own. McKinsey's October 2025 analysis found that in many cases a brand's own sites make up only 5 to 10% of the sources AI search references, and Ahrefs' studies of 75,000 brands found branded web mentions correlate with AI visibility far more strongly than backlinks.

  • Ahrefs, May 2025, AI Overviews: branded web mentions correlated at 0.664 with AI Overview visibility, against 0.218 for backlinks.
  • Ahrefs, December 12, 2025, across ChatGPT, AI Mode and AI Overviews: YouTube mentions showed the strongest correlation, about 0.737. Branded web mentions came in at 0.66 to 0.71, Domain Rating at roughly 0.27 to 0.33, and backlinks were very weak. A site's page count barely registered, at about 0.194.

Ahrefs is explicit that "correlation isn't causation," and the studies cover brands that are already established. Ryan Law, Director of Content Marketing at Ahrefs, summed up the content side in the December 2025 study: "It's not just a content creation arms race."

Rand Fishkin, co-founder of SparkToro, made the training-data version of this argument on October 21, 2024: "The currency of large language models is not links. The currency of large language models is mentions (specifically, words that appear frequently near other words) across the training data." That predates most search-grounded answers, so it describes one of two paths. A model answers from what it learned in training and from what it retrieves at answer time, and mentions feed both.

Google adds a warning: "seeking inauthentic 'mentions' across the web isn't as helpful as it might seem," because its core systems focus on high-quality content and its spam systems apply to AI features too.

So the real question is where to earn mentions, and the answer is the domains the engines already cite for your prompts. Pew found Wikipedia, YouTube and Reddit together made up 15% of the sources in Google's AI summaries. For a B2B SaaS brand, the cited list often includes software review sites, comparison articles, YouTube explainers, Reddit threads and trade publications. Pull your own list from your prompt runs before deciding where to spend, and see AI visibility strategies for the outreach side.

How do you measure AI visibility, and what can't you measure?

Start with the free first-party reports: Search Console's Generative AI performance report for Google's AI features and Bing Webmaster Tools' AI Performance report for Copilot and Bing. Add GA4 referral traffic from AI assistants as a floor, and track brand visibility as a percentage across many prompt runs, never as a rank.

  • Google Search Console: the Generative AI performance reports, introduced June 3, 2026 and available to all sites by August 31, 2026, show impressions from AI Overviews, AI Mode and Discover's AI features. The news post on the report covers how to read it.
  • Bing Webmaster Tools: AI Performance opened in public preview on February 10, 2026, with total citations, average cited pages, grounding queries and page-level citation activity. On June 16, 2026, Microsoft added Intents, Topics, Citation Share and Compare. Citation Share is your site's percentage of all citations shown for a grounding query, and Microsoft calls it an observational metric, not a ranking.
  • GA4: referral sessions from chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com and claude.ai. Treat this as a floor, because clicks from AI Overviews and AI Mode arrive as google / organic and can't be separated.
  • Paid tools: Semrush, in its SEO and AI Search plans, and Ahrefs, through Brand Radar, sell AI visibility tracking. Google cautions that "no third-party tool has access to our internal ranking or AI systems," so use them for sampled prompts, not as a window into Google.

The hard limit is variance. In SparkToro's January 2026 study, 600 volunteers ran 12 prompts through ChatGPT, Claude and Google's AI a combined 2,961 times. ChatGPT and Google's AI returned the same brand list less than once in 100 runs, and the same list in the same order less than once in 1,000. Fishkin's advice was to run a prompt 60 to 100 times before trusting a picture of what an engine recommends, so a single screenshot of an answer proves nothing.

That makes the metric a visibility percentage: the share of runs, across a fixed prompt set, in which your brand is mentioned or your page is cited. Outside Google and Microsoft, you measure through referral traffic and sampled prompts.

Should you block AI crawlers or opt out of AI Overviews?

Allow the retrieval bots. OpenAI says sites that opt out of OAI-SearchBot won't be shown in ChatGPT search answers, and Anthropic says blocking Claude-SearchBot may reduce a site's visibility in Claude's search results. Training crawlers are a separate business decision. Google has let any site opt out of its AI features since August 31, 2026, but an excluded site loses its links and traffic there while competitors' content stays.

  • OpenAI: OAI-SearchBot surfaces sites in ChatGPT search, GPTBot collects training data, and ChatGPT-User handles user-requested fetches, where OpenAI says "robots.txt rules may not apply."
  • Perplexity: PerplexityBot builds its search results and "is not used to crawl content for AI foundation models." Perplexity-User handles user requests and generally ignores robots.txt.
  • Anthropic: Claude-SearchBot indexes for search, ClaudeBot collects training data and Claude-User fetches pages users ask about.
  • Google: Rules addressed to Googlebot affect Google Search, "including Discover and all Google Search features," and products such as Google Images and Google News. Google-Extended controls whether content can train Gemini models and ground answers in the Gemini app and Vertex AI, and Google says it "does not impact a site's inclusion in Google Search."

Google's Search generative AI control, under Settings in Search Console, offers include (the default) or exclude for AI Overviews, AI Mode and generative AI features in Discover. Google's help page is plain about the cost: an excluded site gets no traffic or impressions from those features, and "content from other sites will still be available in those features, and it may appear similar to yours." The control doesn't affect training, which is Google-Extended's job. Nosnippet is blunter still, because it also strips the snippet from regular results.

For a B2B brand that wants to be recommended, I'd allow the retrieval bots, keep the generative AI control on include and make the training call separately. A robots.txt along these lines does that, with the training rules left commented out:

# Search and answer engines: allow
User-agent: Googlebot
Allow: /

User-agent: Bingbot
Allow: /

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

# Training crawlers: uncomment only if you decide to block training
# User-agent: GPTBot
# Disallow: /

# User-agent: ClaudeBot
# Disallow: /

Uncommenting the training lines keeps future content out of those models' training, and it also takes you out of what the models learn about your category. Leave Google-Extended open if you want to be grounded in Gemini app answers, and check your CDN, because a bot-protection default can block crawlers that robots.txt allows.

What should you do in the first 90 days?

Spend the first month on access and a baseline, the second on rewriting the pages that matter and the third on earning mentions and re-measuring. The order matters, because no rewrite helps a page an engine can't index or crawl.

  1. Audit indexing on Google and Bing. Check Search Console's page indexing report and Bing Webmaster Tools for the pages you most want cited, and fix anything crawled but not indexed.
  2. Check crawler access. Review robots.txt, CDN and firewall bot settings, meta robots and nosnippet tags, and confirm the Search generative AI control is set to include.
  3. Fix a prompt set and take a baseline. Write 30 to 50 questions your buyers ask, run each one repeatedly in ChatGPT, Google's AI Mode, Perplexity and Copilot, and record your mention rate and the domains cited. Save the Search Console and Bing reports from the same week.
  4. Rewrite your top 10 pages answer-first, with question headings, a direct answer at the top of each section, dated and linked statistics, named quotes and tables.
  5. Earn mentions on the domains the engines cite. Use the cited-domain list from step 3 to choose review sites, publications, YouTube channels and communities, and give them something worth citing.
  6. Re-measure at day 90 with the same prompts and reports, and compare visibility percentage, citations and AI referrals with the baseline.

Steps 1 to 3 are where search and AI visibility work usually starts: a crawl of the site and a look at what AI assistants say about the brand. For how this fits a wider budget, see GEO in a marketing plan, and if you'd like to talk through your own site, book a 30-minute call.

Frequently asked questions

What is generative engine optimization (GEO)?

Generative engine optimization is the practice of improving how often AI answer engines such as ChatGPT, Perplexity, Copilot and Google's AI Overviews cite your pages and mention your brand. It builds on SEO. The page has to be indexed, crawlable and quotable, then it has to offer the clearest answer, and the brand has to be mentioned on the sites those engines already trust.

Is GEO different from SEO?

GEO uses the same foundation as SEO, and Google says so directly: optimizing for its AI features "is optimizing for the search experience, and thus still SEO." The difference is the outcome. SEO earns a ranking and a click. GEO earns a citation or a brand mention inside an answer, which gives off-site mentions and quotable passages more weight than they carry in rankings.

Is GEO the same as AEO or "AI SEO"?

GEO and answer engine optimization (AEO) describe the same goal, and Google's guide treats them together. "AI SEO" has two meanings: optimizing for AI answers, which is GEO, and using AI tools to do SEO tasks such as research or drafting. When someone sells "AI SEO," ask which of the two they mean before comparing proposals.

Do I need schema markup or an llms.txt file to show up in AI Overviews or ChatGPT?

No. Google says structured data isn't required for its generative AI features, that there's no special schema.org markup to add, and that Google Search ignores llms.txt files. Schema still helps search engines understand your entities and qualify for rich results, so keep it accurate. For ChatGPT, the gate is OAI-SearchBot access in robots.txt, not a special file.

How can I tell whether Google's AI features or ChatGPT cite my site?

For Google, use the Generative AI performance report in Search Console, which shows impressions from AI Overviews, AI Mode and Discover. For Copilot and Bing, use the AI Performance report in Bing Webmaster Tools. For ChatGPT, check GA4 referrals from chatgpt.com and run a fixed prompt set many times, because single answers vary from run to run.

How long does GEO take to show results?

Access fixes can register quickly: OpenAI says robots.txt changes take about 24 hours to reach its search systems, and Perplexity says up to 24 hours. Content rewrites and earned mentions take longer, because pages have to be recrawled and third-party coverage has to build up. I'd judge them against the baseline at 90 days rather than week to week.

Should I block AI crawlers like GPTBot or Google-Extended?

Block a training crawler only if keeping your content out of model training matters more than being part of what models learn about your category. GPTBot and ClaudeBot are training crawlers, and blocking them doesn't remove you from ChatGPT or Claude search. Google-Extended also controls grounding in the Gemini app, so blocking it costs Gemini answers but not Google Search or AI Overviews.

Can I opt my site out of AI Overviews and AI Mode, and should I?

Yes. Since August 31, 2026, Search Console's Search generative AI control lets any site exclude itself from AI Overviews, AI Mode and Discover's generative AI features. Most brands shouldn't. An excluded site loses those impressions and clicks, Google says content from other sites may still appear and look similar, and the setting doesn't stop AI training.

Sources