Generative engineoptimization (GEO):how AI picks sources

What is generative engine optimization?

Generative engine optimization is the practice of shaping your website and your reputation across the web so that AI tools which write answers, rather than list links, choose your business as a source. A generative engine is any tool that replies in sentences: ChatGPT, Perplexity, Gemini, Microsoft Copilot, and the AI Overviews and AI Mode results inside Google.

The name comes from research. In November 2023, six researchers led by Pranjal Aggarwal posted the paper that coined it, later accepted to KDD 2024, a major data science conference. Their own definition:

"we introduce Generative Engine Optimization (GEO), the first novel paradigm to aid content creators in improving their content visibility in generative engine responses"

Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, "GEO: Generative Engine Optimization", arXiv 2311.09735, read October 19, 2026

You will also hear AEO (answer engine optimization), LLM optimization and AI SEO. They describe overlapping work. Google's position is that it is all one discipline: "optimizing for generative AI search is optimizing for the search experience, and thus still SEO." Bing uses the GEO name directly and calls its new citation report "an early step toward Generative Engine Optimization (GEO) tooling." If you want every term defined in its source's own words, our AI search glossary has them.

How do AI engines choose which sources to cite?

They search before they write. The engine turns the question into one or more web searches, pulls pages from an index, reads them in pieces, and links the sources it leaned on. No engine publishes the exact formula, but each one documents enough of the path to know where you can win.

Two of those steps are documented in the engines' own words. Google says AI Overviews and AI Mode may use a "query fan-out" technique, "issuing multiple related searches across subtopics and data sources." OpenAI's help page says "ChatGPT search typically rewrites your query into one or more targeted queries that it sends those providers." Microsoft describes the reading step: "Assistants like Copilot break content down, a process called parsing, into smaller, structured pieces that can be evaluated for authority and relevance."

That is the whole game in two sentences. Your page has to be in the index each engine searches, and once it is retrieved, the passage that answers the question has to be the clearest and best supported one in the pile. Here is where each engine gets its pages:

One more thing shapes every answer: what the model learned in training. Nobody can edit that directly, and anyone who says they can is selling something. What you can do is make sure the live web describes your business clearly and consistently, because that is what both the search step and future training read.

What did the GEO research actually find?

Rewriting pages to read like a well-sourced reference made them show up more in AI answers, and stuffing keywords did not. The researchers built a test set they called GEO-bench, "a benchmark consisting of 10K queries," and measured how much of each AI answer drew on a given source before and after nine kinds of edits. The headline: "GEO can boost visibility by up to 40% in generative engine responses."

Read the limits before you read the numbers. These are lab results on the authors' own benchmark. The test on Perplexity used uploaded text files rather than live websites, because, in the paper's words, Perplexity "does not allow the user to specify source URLs." So the paper tells you which direction to push, not what will happen to your site. The direction is useful: write like a source. Name where facts come from, quote people with standing, give numbers with their origin and date, and say it plainly.

For a deeper look at which tactics hold up across 30 studies, see our AI visibility evidence review.

GEO vs SEO: what changes and what doesn't?

The foundation does not change: crawlable pages, a clean index, helpful content and real authority still do most of the work. Google says unique, compelling and useful content "will likely influence your website's presence in generative AI search in the long run more than any of the other suggestions" in its guide. What changes is what counts as a win, and how much of the work happens off your site.

Is GEO replacing SEO? No. Every engine on this page still starts with a search index, so a site that is weak in classic search starts behind in AI answers too.

How does PELORA do generative engine optimization?

We work in six steps, in this order, because each one depends on the one before it.

1. Open the doors

We confirm that robots.txt and your host or firewall let in Googlebot, Bingbot, OAI-SearchBot and PerplexityBot, that key pages are indexed with snippets allowed, and that no page you want cited carries NOARCHIVE or NOCACHE. Bing says plainly that "NOARCHIVE prevents content from being used in Copilot responses and grounding results." We also read the Search Console setting Google's newer guide names, because it says a site "must be included in Search generative AI features in Search Console" to be eligible, and we submit every change through IndexNow.

2. Make the business one clear entity

Same name, phone, website and one-line description on your site, your Google Business Profile, Bing Places and every directory that matters in your field. Organization markup on the site that matches the visible text word for word. Engines cannot recommend a business they cannot tell apart from its neighbors.

3. Answer the questions people actually ask

We map the fan-out for your main questions: cost, location, reviews, safety, comparisons, insurance, timing. Each real question gets a page or a section that answers it in the first 40 to 60 words. Microsoft's advice is direct: "Don't hide important answers in tabs or expandable menus."

4. Write like a source

Citations to primary sources, quotes from named experts, numbers with their origin and date. That is the pattern the GEO paper found worked best, and it is also what Google asks of health content, which "must be highly accurate and consistent with established expert consensus."

5. Earn mentions where engines look

Directories, review platforms, professional associations, local press, podcasts and credible lists. Reviews are asked for the same way from every customer, never paid for or filtered, in line with Google's review policy and the FTC's rule on consumer reviews.

6. Measure each engine separately

Each engine draws on a different pool of sources, so one blended "AI score" hides what is happening. We report per engine, every month, with the method written down.

What does GEO look like for a healthcare business?

It looks like the method above with a higher bar for proof, because Google holds health content to a higher standard and families check it most closely. Here is how it plays out for the three kinds of businesses we serve.

We have measured these gaps ourselves. When we checked a random sample of 180 licensed California treatment facilities against the federal government's own locator, only 65, or 36%, could be confidently matched to a listing (PELORA study, published August 12, 2026). When we read the robots.txt files of supplement brands, not one of the 28 we could read named a single AI crawler (PELORA study of 30 brands). A source an engine cannot find or match cannot be cited.

What can't GEO do?

It cannot guarantee a citation, and nobody honest will promise one. OpenAI's help page says "Placement is not guaranteed." Bing's guidelines say "GEO does not guarantee grounding or citations in AI experiences." A few other things to know before you hire anyone:

• No one trains ChatGPT for you. No agency can train a model on your business or place you inside its answers.
• No special file unlocks AI answers. Google: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them."
• Mass-produced pages backfire. Google's spam policy defines scaled content abuse as "when many pages are generated for the primary purpose of manipulating search rankings and not helping users."
• Tricks aimed at the model backfire. Bing warns that content "designed to manipulate or interfere with language models" may lead to "reduced visibility or removal from search experiences."

How do you measure generative engine optimization?

You measure each engine on its own, with the platform reports where they exist and a fixed set of questions where they do not. Google's guide points to "the Generative AI performance report in Search Console." Bing's AI Performance report shows "how often your content is cited in generative answers." ChatGPT referral visits arrive labeled chatgpt.com, so they show up in your analytics; that is how we see ours in Google Analytics.

For the engines with no report, we run the same set of buyer questions several times each month and record three separate things: was the business mentioned, was its site cited as a source, and was it recommended. Those three numbers move differently, so we never blend them into one. Then we tie it back to what matters, booked calls and patients, by source. Our own booking form is how we know this channel is real: of the 27 prospects who answered "How did you hear about us?" between September 3 and October 19, 2026, seven named ChatGPT or another AI assistant (PELORA's Calendly records).

What does GEO cost?

Search and AI visibility work is included in every PELORA retainer, and retainers start at $3,500 a month, plus a one-time setup fee. If you want to go further than your tier includes, our AEO, GEO and SEO program runs on its own from $2,000 to $4,000 a month. The exact work is set on the call and written into your proposal before you sign. Every price is on our pricing page.