Generative engine optimization (GEO) is the work of getting your business named and cited in answers written by AI assistants such as ChatGPT, Perplexity, Gemini and Google AI Overviews. The term comes from a 2023 paper by researchers at Princeton, Georgia Tech, the Allen Institute for AI and IIT Delhi, who tested which changes to a web page made generative engines use it more. Their best methods, adding citations, quotations and statistics, improved a page's visibility in generated answers by 30 to 40 percent on their benchmark. Keyword stuffing did nothing, and on Perplexity made things worse.
That paper is a useful place to start because it separates GEO from the rebranding of SEO that fills most articles on the subject. Some of what worked for Google still works. Some of it does not matter at all. This guide covers what actually changes when a buyer asks an assistant instead of typing into a search box, and what a small team should do about it.
We make Seen42, an AI visibility tool, so read our product notes with that in mind. Everything else in this guide works without it.
What changes when the answer is written, not listed
A search engine returns ten links and lets the buyer choose. An assistant reads a handful of pages, writes one answer, and names the brands it found in them. Four things follow from that.
Fewer brands get named. In a check we ran on 2026-10-20, "What are the best AI visibility tools for small businesses?" produced answers that each named roughly five to seven tools. Position eleven on Google still gets some clicks. Not being in the answer gets nothing.
The sources are chosen per answer, not per site. Google's documentation describes its AI features issuing several related searches, a "query fan-out", and picking supporting pages from the results. Perplexity describes searching the web in real time and citing what it reads. The page that gets cited is the one that answers the specific question, which is often not your homepage.
Third-party pages carry more weight than you expect. For the tools question above, the assistants cited vendor blog roundups, review sites and comparison articles. Not one cited a vendor's own pricing page. If you want to be named, you need to be on the pages assistants already trust for your prompts, as well as on your own.
Answers vary. Ask the same question twice and the brands change. GEO is measured in trends over repeated checks, not in a single screenshot.
What stays the same
Google says there are "no additional requirements to appear in AI Overviews or AI Mode", no new files or markup, and that the usual crawling and indexing rules apply. The same page quality that earned a snippet earns a place in an overview. OpenAI and Perplexity both publish search crawlers that respect robots.txt, so being crawlable is still the first gate.
The research agrees. The GEO paper found that making a page "authoritative" in tone did nothing measurable, while adding verifiable specifics did a lot. Specific, sourced, readable pages were good SEO before and they are good GEO now.
What the research says works
The GEO paper tested nine methods on a benchmark of 10,000 queries. In order of effect:
- Cite sources, add quotations, add statistics: the strongest group, with a 30 to 40 percent relative improvement on the paper's position-adjusted word count metric.
- Make the text easier to understand, improve fluency: a 15 to 30 percent improvement.
- Add technical terms, add unique words: smaller gains.
- Sound authoritative: no significant improvement.
- Keyword stuffing: little or no improvement overall, and about 10 percent worse than the unoptimized page on Perplexity.
The paper also found that lower-ranked pages gained the most. Adding citations raised the visibility of the fifth-ranked source in its tests by more than 100 percent while the top-ranked source lost ground. For a small company that will never be the top result, that is the most encouraging finding in the field.
A GEO plan for a team without a marketing department
- Write down 10 to 20 buyer prompts: category questions, comparisons and problems, in the words a buyer would type. Our guide to finding the questions buyers ask AI walks through this.
- Ask the assistants and record who gets named and which pages get cited. That list of cited pages is your plan.
- Check that OAI-SearchBot, PerplexityBot, Claude-SearchBot and Googlebot can reach your site. Our AI crawler guide has the details.
- For each prompt where a competitor is named and you are not, publish one page that answers it: a short answer at the top, specific facts with dates and sources, a table where the question is a comparison, and a plain statement of who the page is for.
- Get your facts onto the third-party pages that were cited: review profiles, directory listings, comparison articles that accept corrections.
- Keep every fact about you identical everywhere: name, prices, plans, integrations, who you serve.
- Re-check the same prompts every week and read the trend against a few control prompts you did nothing for.
What to ignore
Ignore anyone selling a guaranteed ranking in ChatGPT; assistants choose their own sources and change their minds. Ignore advice to stuff the words "best" and "top" into headings; the one study that measured it found it hurts. Treat llms.txt as a cheap, optional courtesy rather than a strategy; no major assistant has said it reads the file, and our llms.txt guide explains the current state.
Where Seen42 fits
Seen42 checks the prompts your buyers ask on ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews, shows who is named and cited instead of you, and builds and hosts comparison, facts and guide pages from your own facts. Paid plans start at $39 a month for 15 prompts; the Free plan checks 10 prompts once on ChatGPT and Gemini. It does the plan above as a loop: scan, build the page, publish it after your approval, and re-check the same prompts so you can see whether anything changed.
