AI SEO (GEO): What It Is, What Actually Works, and How to Measure It in 2026

If you search for advice on “AI SEO,” “GEO,” “AEO,” “LLM optimization” or “ChatGPT SEO,” you can find a confident answer for almost anything. Add an llms.txt file. Rewrite every paragraph into bite-sized chunks. Publish hundreds of FAQs. Get mentioned on Reddit. Add schema. Buy a tool that checks 500 prompts a day.

Some of those ideas are useful in the right context. Some are ordinary SEO with a new label. Some are unsupported. And some advice that sounded plausible in 2024 is already contradicted by what Google, Microsoft and OpenAI now publish themselves.

The key word is more likely. There is no guaranteed ChatGPT ranking, no universal GEO checklist and no single position to optimize. AI answers can change with the prompt, model, location, user context, source availability and product updates. That makes GEO measurable, but not in exactly the same way as traditional SEO.

Where the term GEO came from

The term “Generative Engine Optimization” was formalized by researchers from Princeton University, Georgia Tech, the Allen Institute for AI and IIT Delhi in the paper GEO: Generative Engine Optimization, first posted in 2023 and later published at KDD 2024.

The paper describes generative engines as systems that retrieve sources and then synthesize an answer from them. Instead of competing only for position one through ten, publishers are competing to become useful source material inside the generated answer itself.

In the researchers’ GEO-bench experiments, tactics such as adding relevant statistics, quotations and citations improved source visibility, while keyword stuffing did not perform well. The paper reported visibility gains of up to 40% on its benchmark and up to 37% in a Perplexity experiment. Those are experimental results, not a promise that adding a statistic to a service page will increase your ChatGPT mentions by 40%.

GEO vs. SEO: different output, shared foundation

Our SEO services address the crawlability, useful content, and local visibility that support both traditional search and AI discovery.

Traditional SEO usually asks: can this page be crawled, indexed and ranked for a query? GEO adds another question: can an AI system retrieve this information and confidently use it in an answer?

That sounds like an entirely new discipline, but the official guidance is much less dramatic. Google says the same foundational SEO practices apply to AI Overviews and AI Mode, with no special technical requirement beyond being indexable and eligible to appear in Search. Bing’s webmaster guidelines say SEO fundamentals also support grounding results and citations in Copilot and other AI experiences.

So the most useful mental model is not “SEO is dead.” It is:

How AI search actually finds information

There is no single architecture behind every AI answer, but current search products increasingly use retrieval and grounding. A user asks a question, the system reformulates or expands that question, retrieves sources, and then generates an answer grounded in some of those sources.

Google explicitly says AI Overviews and AI Mode can use a “query fan-out” technique, issuing multiple related searches across subtopics and data sources. Bing describes grounding as a core layer connecting AI-generated answers to web information. OpenAI says public websites can appear in ChatGPT search and tells publishers to allow OAI-SearchBot if they want their content discoverable for summaries and snippets.

That leads to a practical principle: you are not optimizing only for the exact words a user types. You are optimizing the set of facts, pages and third-party sources an AI system might retrieve while trying to answer the broader question.

For local businesses, that retrieval process becomes especially important because the answer can vary with location, third-party business data, and the specific service a customer asks about. We break that down in How Does ChatGPT Choose Which Local Businesses to Recommend?

Two companion guides cover the practical and evidentiary sides of the same problem: How to Get Your Business Recommended by ChatGPT in 2026 and AI Trust Signals: What Makes ChatGPT Trust and Recommend a Business?

Why this matters: AI answers change click behavior

Pew Research Center analyzed the March 2025 browsing behavior of 900 U.S. adults. In that panel, users clicked a traditional Google result on 15% of visits when no AI summary appeared, but only 8% when an AI summary did appear. That is about a 47% lower traditional-result click rate on a relative basis.

Traditional Google result clicks when an AI summary appears

Share of observed Google search visits that led to a traditional result click, Pew Research Center browsing panel, March 2025.

No AI summary

15%

AI summary shown

8%

Tacklebox chart. Source data: Pew Research Center, “Google users are less likely to click on links when an AI summary appears in the results,” published July 22, 2025. This is observational panel data, not a causal experiment. The 47% figure is our calculation from Pew’s 15% and 8% reported rates.

Pew also found that only 1% of visits to pages with an AI summary resulted in a click on a cited source inside that summary. Users ended their browsing session after 26% of search pages with an AI summary, compared with 16% of pages without one.

At the same time, Google says AI Overviews are increasing search usage for the types of queries where they appear. Those two things can both be true: AI can increase the number and complexity of searches while decreasing the probability that any one search produces a conventional website click.

For marketers, that means visibility cannot be measured only by organic sessions. Being the source, brand or business included in the answer can have value even when the user never clicks the citation.

AI answers show up most on the kinds of questions businesses care about

Pew found AI summaries on about 18% of the Google searches in its March 2025 sample, but they were much more common on longer and question-style searches. AI summaries appeared on 53% of searches with ten words or more and on 60% of searches beginning with words such as who, what, when or why.

Those longer prompts are important because they often look more like real buying questions:

• Who is the best commercial roofer near me for a flat roof?
• What Nashville SEO agency works with local home-service businesses?
• Which dentist in my area offers same-day crowns and has strong reviews?
• What pressure washing company is best for a large commercial property?

Traditional keyword research is still useful, but GEO requires thinking in questions, comparisons, attributes and decision criteria, not just keywords.

What actually appears to work for GEO

The field is moving too quickly for a permanent ranking-factor list. But there is now enough first-party documentation to separate durable work from speculation.

1. Make sure the content can actually be crawled and indexed

This is boring and essential. Google says a page needs to be indexed and eligible to appear with a snippet to be eligible as a supporting link in AI features. Bing ties AI grounding eligibility to the same crawl and indexing foundation used by search. OpenAI tells publishers who want to appear in ChatGPT search to allow OAI-SearchBot and make sure their host or CDN is not blocking its published IP ranges.

Before “optimizing for AI,” check robots.txt, noindex directives, canonical tags, internal links, sitemaps, rendering and bot protection. If the retrieval system cannot reliably access the page, the rest is decoration.

2. Publish information that is worth retrieving

Google’s current generative-search guidance emphasizes unique, useful, non-commodity content. Bing similarly recommends original, differentiated material and warns against large-scale low-value generated content.

For a business, that means the highest-value GEO content is usually not another generic “what is X?” article. It is content only that business can credibly publish: original project data, before-and-after details, service limitations, comparison criteria, methodology, expert commentary, local knowledge, proprietary research and clear answers to real customer questions.

3. Support claims with evidence

This is one area where the original GEO paper and current Bing guidance point in the same direction. The GEO paper found strong benchmark performance from statistics, quotations and citations. Bing explicitly recommends supporting claims with examples, data and cited sources so content is clear and trustworthy when reused in AI-generated answers.

That does not mean adding random statistics for the sake of it. The useful pattern is simple: make a claim, provide evidence, cite the source, and explain why it matters.

4. Structure pages so the answer is easy to extract

Descriptive headings, concise sections, lists, tables and FAQs are useful because they make information easier for both humans and retrieval systems to parse. Bing specifically recommends clear structure, and Google continues to recommend conventional people-first page organization.

This is different from the more extreme “chunk every paragraph into 50 words for LLMs” advice. Google’s 2026 guidance explicitly says you can ignore artificial chunking tactics for Google Search. Write for people first; make the document easy to navigate second.

5. Build a consistent entity across the public web

AI systems do not operate only on your homepage. They can retrieve business directories, review platforms, news coverage, community discussions, social/video pages, reference sources and other sites.

Pew found Wikipedia, YouTube and Reddit were the three most frequently cited domains across both AI summaries and standard Google results in its sample; together they represented 15% of sources cited in the AI summaries studied.

That is not proof that “posting on Reddit makes ChatGPT rank you.” It is evidence that public third-party sources are part of the information environment AI search systems use. For a local business, consistency across your website, Google Business Profile, major directories, review sites, local press and relevant industry sources makes the business easier to verify.

6. Earn real mentions instead of manufacturing them

Third-party mentions can be valuable because they provide independent context about a company. But inauthentic mention campaigns are exactly the kind of shortcut that search systems have spent decades learning to discount.

Google’s current AI optimization guide specifically warns site owners not to pursue inauthentic mentions as an “AEO/GEO hack.” The durable version is the same one that has always worked: do something worth covering, contribute expertise, build partnerships, publish useful data and earn reviews from real customers.

7. Keep normal SEO strong

There is no clean line where SEO stops and GEO starts. If your important service pages are weak, your site is not indexed properly, your brand has almost no external authority and your local business information is inconsistent, an “AI optimization” layer is not going to rescue it.

Google says the SEO foundation carries directly into generative AI search. Bing says the same. GEO should be treated as an extension of search strategy, not an excuse to skip it.

What we would not spend much time on

Tactic	Our view	Why
llms.txt for Google visibility	Low priority	Google explicitly says it does not use llms.txt as a special signal for generative Search.
Keyword stuffing	Avoid	The original GEO benchmark found keyword stuffing underperformed; search-engine spam policies already discourage it.
“AI schema” or special GEO markup	Unproven	Google says there is no special markup required for AI Overviews or AI Mode. Use accurate structured data where it genuinely describes the page.
Mass AI content	Risky	Commodity content adds little unique evidence and can create indexing-quality problems.
Fake Reddit / forum mentions	Avoid	Public discussions can matter, but manufactured recommendations are not a durable authority strategy.
Original data, proof and expert detail	Worth doing	Supported by research, official guidance and common sense: it gives retrieval systems something specific to cite.

How to measure GEO without fooling yourself

This is where a lot of AI-visibility reporting gets shaky. Running one prompt in ChatGPT and taking a screenshot is not measurement. Neither is treating one model’s answer as a permanent ranking.

We prefer to measure four layers.

1. Prompt coverage

Create a fixed set of real customer questions and buying prompts. Run the same prompts on a schedule across the AI surfaces that matter to the business. Record whether the brand appears, whether it is recommended, what competitors appear, and which sources are cited.

Do not obsess over hundreds of random prompts. A smaller set tied to actual commercial intent is usually more useful.

2. Citation visibility

Track which pages and third-party sources are being cited. Bing now exposes this directly in its AI Performance report, including total citations, cited pages, grounding queries and trends over time.

Google also rolled out dedicated Generative AI performance reports in Search Console in 2026, with visibility into impressions, pages, countries, devices and dates for generative AI features. That is significantly better evidence than guessing from a third-party score.

3. Referral traffic

Track traffic from AI products in analytics. OpenAI says ChatGPT search referral links automatically include utm_source=chatgpt.com, which makes identifiable ChatGPT search traffic easier to segment.

Referral data will never capture the full value of an AI mention because many users will see a recommendation and later navigate directly, call, search the brand or convert offline. But it is still a useful signal.

4. Business outcomes

The final question is the same as every other marketing channel: did it create qualified demand? Add “How did you hear about us?” to lead forms or intake calls. Watch branded search. Track calls, forms, booked work and revenue where possible.

AI visibility is interesting. AI visibility that generates customers is useful.

GEO for local businesses

Local GEO deserves its own treatment because a local recommendation question is different from an informational query. If someone asks an AI assistant for a roofer, dentist, restaurant, landscaper or attorney nearby, the answer may need to establish location, category, reputation, service fit and trust all at once.

For a local business, our practical stack is:

1. Get the core entity right. Name, address/service area, phone, categories, services and website information should be consistent.
2. Build strong service pages. Explain exactly what you do, who it is for, where you do it, what it costs when appropriate, and what makes the service different.
3. Keep Google Business Profile complete. Categories, services, reviews, photos and business information still matter to the broader local search ecosystem.
4. Earn third-party proof. Reviews, local coverage, trade associations, suppliers, community organizations, directories and relevant editorial mentions help establish that the business exists and is trusted.
5. Answer recommendation prompts. Build content around the questions customers ask before choosing: comparisons, project types, neighborhoods served, constraints, timelines and costs.
6. Measure the answers. Track a fixed set of local commercial prompts over time and compare your presence with real competitors.

Where Reddit fits into AI SEO

Reddit is interesting for two separate reasons.

First, it is a source. Pew found Reddit among the most frequently linked domains in the Google AI summaries it studied. Public community conversations can contain product experiences, local recommendations, comparisons and language that does not exist on polished corporate websites.

Second, Reddit is one of the better places to watch practitioners argue in public about what they are actually seeing. That does not make every Reddit comment reliable. It makes the communities useful for discovering experiments, edge cases, tracking tools and questions worth testing yourself.

For GEO specifically, we would watch:

Use communities to learn and participate honestly. Do not create fake recommendation threads about your own company. Besides being unethical, that strategy is fragile and difficult to scale without eventually looking exactly like what it is.

The resources we trust most

Our current GEO playbook

If we were starting from zero for a business today, we would not begin with a special AI file or a giant prompt-tracking subscription. We would work in this order:

1. Fix crawlability, indexation and technical SEO.
2. Define the real customer questions and commercial prompts that matter.
3. Build or improve the pages that answer those questions with specific, original information.
4. Add evidence: data, examples, expert detail, sources, project proof and useful comparisons.
5. Strengthen the business’s presence across authoritative third-party sources.
6. Make structured business information accurate and consistent.
7. Track a fixed prompt set, cited sources and official AI-search visibility data.
8. Measure leads and revenue instead of declaring victory because the brand appeared once in ChatGPT.

That is less exciting than promising a secret GEO hack. It is also much closer to what the available evidence supports.

AI SEO and GEO FAQs

Is GEO the same as AI SEO?

In practice, people use AI SEO, GEO, AEO and LLM optimization with a lot of overlap. We use “AI SEO” as the broad category and “GEO” for the specific goal of improving visibility, citation and recommendation inside generative answers.

Does GEO replace traditional SEO?

No. Google and Bing both explicitly connect AI visibility to traditional crawling, indexing, quality and relevance foundations. GEO extends SEO into a new answer format; it does not eliminate the need to rank and be understood by search systems.

Do I need llms.txt?

Not for Google Search. Google’s current documentation explicitly says llms.txt is not needed for its generative AI search features. Other systems could adopt new conventions over time, so this is something to revisit rather than treat as a universal rule.

Does schema help with ChatGPT or AI Overviews?

Accurate structured data can help machines understand entities and page content, and it remains useful in ordinary SEO. But Google says there is no special structured-data requirement for AI Overviews or AI Mode. We would treat schema as good technical hygiene, not a guaranteed GEO lever.

How do I rank number one in ChatGPT?

That is the wrong model. Generative answers are not a stable ten-blue-link ranking. Measure whether your brand is mentioned or recommended across a representative set of prompts, which sources are cited, how that changes over time and whether it produces business.

Sources

1. Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024.
2. Google Search Central, Guide to optimizing for generative AI features on Google Search.
3. Google Search Central, AI features and your website.
4. Google Search Central, Introducing Search Generative AI performance reports in Search Console.
5. Bing Webmaster Tools, AI Performance and current webmaster guidelines.
6. OpenAI, Publishers and Developers FAQ.
7. Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results, July 22, 2025.