AI Search and GEO · Guide
Generative engine optimization (GEO): a practical guide for 2026
Generative engine optimization (GEO) is the work of getting your pages read and cited, and your brand named, in answers from ChatGPT, Perplexity, Claude and Google's AI Overviews and AI Mode. In practice it is SEO on a different list of queries, plus work to be mentioned on the pages those engines read. Google's own AI optimization guide (May 2026) says classic SEO is still the foundation.
What is generative engine optimization?
GEO is optimizing content so that AI systems which write answers, rather than list links, pick it as a source and repeat its facts. The term comes from a research paper, "GEO: Generative Engine Optimization" by Aggarwal and colleagues, first posted in November 2023 and accepted to KDD 2024, which reported visibility gains of up to 40% in generative engine responses in its tests.
Since then the word has been stretched to cover almost anything with "AI" in it. A useful working definition for a small business has three parts:
- Be retrievable. The search bots that fetch pages for answers can reach your pages and read the text.
- Be the right source. You have pages that already contain the shape of the answer: named options, prices with dates, trade-offs.
- Be named by others. Pages on other sites that the engines read mention you.
The first two are SEO you already know. The third is where most of the new effort goes. If you want the vocabulary sorted out first, read GEO vs AEO vs SEO and what answer engine optimization means.
What GEO is not
GEO is not a separate discipline with its own secret files. Google's AI optimization guide tells site owners they can ignore "chunking" content, creating "unnecessary AI text files (like llms.txt)" and "pursuing inauthentic mentions". The common myths, checked against primary sources:
| Claim you will hear | What the source says (checked 16 Nov 2026) |
|---|---|
| You need an llms.txt file to show up in AI answers | Google Search "doesn't use" llms.txt; it "will neither harm nor help" visibility (Google AI optimization guide). Other systems may read it. |
| You need special schema for AI Overviews | "There's also no special schema.org structured data that you need to add" (Google AI features page) |
| FAQ schema gets you a rich result | FAQ rich results stopped showing on 7 May 2026 and the doc was removed 15 Jun 2026 (Google Search updates) |
| Blocking GPTBot keeps you out of ChatGPT search | GPTBot is for training. OAI-SearchBot controls ChatGPT search (OpenAI bots doc) |
| Google-Extended opts you out of AI Overviews | It does not. Googlebot rules and the new Search generative AI control in Search Console do |
| You should publish a page for every AI prompt | Google says making pages for fan-out variations primarily to manipulate generative AI responses violates its scaled content abuse policy |
How an AI answer picks its sources
An engine that searches does three things: rewrites the question into several short searches, retrieves pages for each, then writes an answer from what it read and links some of those pages. Your page has to survive all three steps.
Retrieval bots versus training bots. OAI-SearchBot, Claude-SearchBot, PerplexityBot and Googlebot fetch pages so they can appear in answers. GPTBot, ClaudeBot and Google-Extended relate to model training. Blocking a retrieval bot removes you from that engine's answers; OpenAI's docs say sites opted out of OAI-SearchBot "will not be shown in ChatGPT search answers". The full robots.txt and curl checks are in the technical SEO for AI search checklist.
Fan-out. Google describes AI Overviews and AI Mode "issuing multiple related searches across subtopics and data sources". Its own example: a question about a weedy lawn fans out into searches such as "best herbicides for lawns". Query fan-out explained covers what is measured and what is not.
Rotation. The same prompt does not return the same sources every time. That is why GEO measurement counts how often you are named across repeated runs, not a single position.
Ranking in Google helps, but less than you would think
Ranking in classic search raises your odds of being cited, but the overlap is partial and varies by engine. Four studies, each with a different method, show the spread:
| Study (publisher, date) | What was measured | Result |
|---|---|---|
| Ahrefs, Aug 2025 | Links cited by ChatGPT, Gemini, Copilot that also rank in Google's top 10 for the same prompt | 12%; Perplexity 28.6% |
| Semrush, Jul 2025 | Position of ChatGPT-cited pages in Google | Rank 21 or lower "almost 90% of the time" |
| BrightEdge, Sep 2025 | AI Overview citations overlapping organic rankings | Grew from 32.3% to 54.5% over 16 months |
| Ahrefs, Mar 2026 | AI Overview URLs also in the first 10 result blocks | 37.9% (method changed since a 76% reading in Jul 2025) |
Read this as: Perplexity and AI Overviews lean on search rankings more than ChatGPT does, and no single number is settled. Engine by engine tactics are in how to get cited by ChatGPT, how to rank in ChatGPT search, how to get cited by Perplexity and how to rank in Google AI Overviews.
The six page types answers are built from
AI answers to buying questions are assembled from a small set of page shapes. The GEO guide by Nicholas Dulait (13 Nov 2026) names six, in the order to build them for AI answers:
| Page type | Query it answers | Must contain |
|---|---|---|
| Comparison | "X vs Y" | Both names in the H1, verdict in 3 sentences, table, a row you lose |
| Alternative | "X alternatives" | 5 to 8 named options with dated prices, you honestly placed |
| Integration | "Does X work with Shopify" | Yes or no in line one, numbered setup |
| Constraint | "tools without a credit card" | Binary answer; skip it if you do not meet it |
| Feature | "tools with an API" | Only features you genuinely lead on |
| Problem | "how to automate X" | One task, numbered steps, pitfalls |
Each type is unpacked with examples in the six page types AI answers are built from.
How to write a page an answer can lift
Write so a paragraph can be quoted without the rest of the page. One question per H2, the answer in the two or three sentences right under it, then the detail. Use names, not categories ("Ahrefs and Semrush", not "the big suites"), numbers with units and a date ("$49 a month as of November 2026"), and tables when you compare more than two things. Say what you do not do.
Google's guide adds a quality bar: content that is "non-commodity", meaning it "provides unique expert or experienced takes that go beyond common knowledge". A page that restates what any model could write has nothing for an engine to cite. The full method with before and after examples is in how to write answer-first content, and the whole checklist is in the AI search optimization checklist.
Being named by other sites
Off-site mentions probably matter more in GEO than in classic SEO. Ahrefs found brand web mentions had the strongest correlation (0.664) with AI Overview brand visibility across 75,000 brands, against 0.218 for backlinks (May 2025, correlation only). In its study of ChatGPT's top 1,000 cited pages (October 2025), roughly two thirds were "effectively off-limits to traditional outreach tactics", so you will not email your way onto most of them.
What a small site can do: get onto the "best X" lists and review pages that already rank for your category, publish honest comparisons that others link to, and appear where your buyers talk. What to avoid: buying links that pass ranking credit, which Google's link spam policy forbids unless they carry rel="sponsored" or rel="nofollow". Postbing's position on this is on the safe SEO page.
There is one more trap. In Lily Ray's study of AI Overviews (June 2026), when Google cited a brand's own self-promoting "best" listicle, it left that brand out of the actual recommendation 69% of the time. Ranking yourself first on your own list can turn your page into a source for your competitors.
Freshness
Fresh pages get cited more often, though the evidence is a correlation. In Ahrefs' October 2025 study, 76.4% of ChatGPT's most cited pages with a detectable date had been updated in the last 30 days (564 dated URLs out of the top 1,000). Put "as of " next to every price and number, and refresh pages older than about 90 days with real changes, not a new date stamp.
How to measure GEO
Measure how often you are named, not a rank. There is no search volume database for prompts, so every tracker samples prompts someone chose. Use four instruments together:
- A weekly prompt log: 10 buyer prompts, logged out, 3 runs each per engine, sources opened. Method and template: track AI citations for free.
- AI referral traffic in GA4 through a custom channel group. Setup and a regex that works with GA4's full match rule: AI search traffic in GA4.
- Search Console: Google counts AI feature traffic in the Performance report and its AI optimization guide points to a Generative AI performance report.
- Ask people. "How did you hear about us?" at signup, with ChatGPT, Perplexity and Google AI as options.
The broader system, including paid trackers and their blind spots, is in the AI visibility tracking guide.
What nobody knows yet
Be skeptical of anyone who sounds certain, including us. Most GEO numbers are vendor studies and small panels:
- Correlations, not causes. Brands mentioned everywhere are often simply good brands.
- Numbers disagree by method. Reddit's share of ChatGPT citations was reported as falling from close to 60% of responses to around 10% (Semrush, 2025) and from 3.8% to 0.5% of citations (Promptwatch data via Semrush, 2026). Both can be true because they measure different things.
- Value of the traffic. Ahrefs measured AI at 0.1% of total referral traffic across about 35,000 sites (March 2025). Semrush estimated the average AI visitor is 4.4 times as valuable as an organic one by conversion rate (July 2025). Small volume, possibly high intent.
- APIs differ from apps. Answers fetched through an API can cite different sources than the app a logged-out person uses.
A 90 day GEO plan for a small site
- Week 1: run the bot and raw HTML checks, add the GA4 channel and the "how did you hear about us" question, and do a first logged-out prompt run.
- Weeks 2 to 6: list your fan-out queries from your best commercial keywords. Publish one comparison or alternative page a week, honestly placed.
- Weeks 6 to 10: find the ranking "best" lists that leave you out and ask to be reviewed. Add integration and constraint pages you genuinely qualify for.
- Weeks 10 to 13: refresh anything with a price or date older than 90 days, re-run the prompt log, and compare named rates. On a young domain, publish about two pages a week and wait roughly two months before judging a change.
You can also use assistants to help with the work itself: see how to use Claude for SEO and how to use ChatGPT for SEO.
Where Postbing fits, and where it does not
Postbing plans, writes and publishes articles to your CMS with meta tags, FAQs and JSON-LD in each one, refreshes underperformers from Search Console data, and runs a weekly AI visibility check that asks your buyer questions through the AI providers' APIs (5 to 40 questions per site per week depending on plan, as of November 2026). It never sells or places backlinks.
It does not do outreach, ranking swaps, video, Reddit or review platforms, which is much of the off-site work above. It does not guarantee citations or rankings, and its API-based checks should be paired with your own logged-out log. Plans and prices are listed on the pricing page linked below.
FAQ
Is GEO different from SEO?
Mostly no. It uses the same foundations (crawlable pages, useful content, links and mentions) aimed at the short searches AI engines run, plus extra work to be named on third-party pages. Google's AI optimization guide says foundational SEO practices remain the base.
Do I need llms.txt for generative engine optimization?
No. Google says Google Search does not use llms.txt and that it neither helps nor harms visibility. It is cheap to publish for other systems, but do not expect ranking or citation gains from it.
How long does GEO take to show results?
Expect weeks to months. Engines cite pages they have already crawled and indexed, and on a young domain it is sensible to wait about two months before judging a change. Track named rates weekly so you can see the trend.
Which AI engine should I optimize for first?
Start with the one your buyers use, which your signup question will tell you. If you rank well in Google already, Perplexity and AI Overviews overlap more with search rankings than ChatGPT does, per Ahrefs' August 2025 study.
Can a tool guarantee AI citations?
No. Sources rotate between runs and engines change often. Google's own guide warns against third-party tools that "promise ranking success".
Sources
- GEO: Generative Engine Optimization (arXiv): origin of the term, 40% visibility result, KDD 2024, accessed 16 November 2026
- Google: Optimizing for generative AI search: foundations, chunking and llms.txt, non-commodity content, fan-out pages and scaled content abuse, Generative AI performance report, accessed 16 November 2026
- Google: AI features and your website: no special requirements or schema, query fan-out, Search Console reporting, accessed 16 November 2026
- Google Search Central updates: FAQ rich result removal dates, accessed 16 November 2026
- OpenAI: Overview of OpenAI crawlers: OAI-SearchBot and GPTBot roles, accessed 16 November 2026
- Google: Spam policies: link spam and qualified links, accessed 16 November 2026
- Ahrefs: AI search overlap: 12% and 28.6% overlap, accessed 16 November 2026
- Ahrefs: AI Overview citations and the top 10: 37.9%, accessed 16 November 2026
- BrightEdge: AIO rank overlap after 16 months: 32.3% to 54.5%, accessed 16 November 2026
- Semrush: AI search SEO traffic study: rank 21+ and 4.4x value, accessed 16 November 2026
- Ahrefs: ChatGPT's most cited pages: two thirds off-limits to outreach, 76.4% freshness, accessed 16 November 2026
- Ahrefs: AI Overview brand correlation: 0.664 and 0.218 correlations, accessed 16 November 2026
- Ahrefs: AI traffic research: 0.1% of referral traffic, accessed 16 November 2026
- Lily Ray: Why calling yourself the best could help competitors: 69% self-listicle result, accessed 16 November 2026
- Semrush: Most cited domains in AI and Semrush: Reddit citations in ChatGPT fall: Reddit share figures, accessed 16 November 2026
- Nicholas Dulait: The KING of GEO (X post): six page types, prompt log method, publishing pace, accessed 16 November 2026
Related reading
Last updated: November 16, 2026. Spotted an error? Email support@postbing.com and we will fix it. Editorial policy.
Try Postbing free
Postbing reads your site, plans a month of topics from what your customers search for, writes in your voice and publishes to WordPress, Shopify, Webflow or Ghost. Your first article is free.
Try Postbing freeNo card needed. No link exchanges, ever.