Generative engine optimization: how to get cited by AI search

Generative engine optimization is the newest name for an old goal: being the source people find when they search. Today the searcher often reads an answer written by Google’s AI Overviews, ChatGPT search, Microsoft Copilot or Perplexity, and sees your business only if that answer cites you. This guide covers how those systems choose sources, what research says moves visibility, and a plan you can run this month.

What is generative engine optimization?

Generative engine optimization is the practice of shaping your content so that AI-generated search answers use it and link to it. The term comes from a research paper by a team from Princeton University and partners, published at the KDD 2024 conference, which defined a generative engine as a system that retrieves web pages and writes one combined answer from them.

Classic SEO tries to rank a page in a list of links. GEO tries to get a page selected as one of the few sources behind an answer. The two overlap heavily, because most AI search tools retrieve from a normal search index first. But the reward is different: a citation inside an answer, not a position on a results page.

You will also hear answer engine optimization (AEO). We treat AEO as the page-level craft of writing passages that assistants can quote, and GEO as the wider job of being visible and trusted across AI search. Our guide to building pages that AI assistants quote covers the writing side in detail.

How do AI search tools decide which pages to cite?

AI search tools cite pages they can retrieve from a search index, that match the question closely, and that state the answer clearly enough to support a sentence in the response. Google says its AI features use retrieval-augmented generation, also called grounding, and query fan-out, which means the model fires off several related searches, pulls pages from the Search index and writes an answer backed by links to them.

Retrieval-augmented generation (RAG) is the same pattern a company chatbot uses to answer from your own files; our explainer on how a chatbot answers from your documents walks through it.

Each platform has its own route in:

• Google AI Overviews and AI Mode: a page must be indexed and eligible to show with a snippet. Google states there are no additional technical requirements and no special markup needed to appear.
• ChatGPT search: OpenAI uses a crawler called OAI-SearchBot. OpenAI says sites that block it will not be shown in ChatGPT search answers, although they can still appear as plain navigational links. GPTBot, the training crawler, is a separate setting.
• Microsoft Copilot and Bing: these draw on the Bing index. Microsoft explains that Copilot parses pages into smaller, structured pieces and evaluates each piece for authority and relevance to the question.

The common thread: if a crawler cannot reach your page, or the answer sits inside an image, a PDF or a script that never renders, clever writing cannot help.

What does the research say actually works?

The original GEO study tested nine ways of rewriting web pages across 10,000 queries and measured how much of the AI answer drew on each source. The clearest finding: adding verifiable substance worked, and old keyword tricks did not. The figures below are the paper’s overall position-adjusted word count scores, taken from Table 1, with the relative change from an unedited baseline of 19.3.

Method tested	Score	Change vs baseline
Adding quotations from relevant sources	27.2	+41%
Adding statistics	25.2	+31%
Improving fluency and readability	24.7	+28%
Citing sources	24.6	+27%
Keyword stuffing	17.7	-8%

Two findings matter most for small and mid-sized businesses. First, keyword stuffing was the only method that made things worse. Second, pages that ranked lower in normal search gained the most. In the paper’s breakdown, citing sources lifted fifth-ranked sources by about 115% while slightly lowering the top-ranked ones. Specific, sourced content gives a smaller site a real chance next to bigger names.

Treat these numbers as direction, not promise: they come from a 2023 to 2024 lab benchmark, and live systems change often. The pattern does match what the platforms now say in their own guidance.

Is GEO different from SEO?

GEO is mostly SEO done well, with a few extra habits. Google is blunt about this: its guide says that from Google Search’s point of view, optimizing for generative AI search is still SEO. The same guide lists things you can safely ignore for Google:

• llms.txt files and special AI markup. Google Search does not use them. They neither help nor harm your Google visibility, though other services may read them.
• Chopping content into tiny chunks. Google says its systems can find the relevant part of a longer page.
• Rewriting everything “for AI”. AI systems understand synonyms, so you do not need every phrasing of a question.
• Chasing inauthentic mentions on forums or low-quality sites. Google says this helps less than it seems, and its AI features rely on the same spam systems as Search.
• Mass-producing pages for every query variant. Google classes this as scaled content abuse.

GEO adds three habits. Write sections that each answer one question, because assistants evaluate sections and readers skim the same way. Keep your business facts identical everywhere AI tools look, from your site to your Google Business Profile and directories. And measure citations, not just clicks.

A 6-step GEO plan you can run this month

This plan assumes an existing website and a few hours a week.

1. List 20 to 30 real buyer questions. Pull them from sales calls, support emails and your Search Console queries. Mix stages: “what is”, “how much does”, “X vs Y” and “best option for [situation]”.
2. Test them in three assistants. Ask each question in ChatGPT, Gemini or Google AI Mode, and Perplexity or Copilot. Record in a spreadsheet whether you are mentioned, who is mentioned instead, and which URLs are cited.
3. Check crawler access. Open your robots.txt and your CDN or firewall settings. Make sure Googlebot, Bingbot and OAI-SearchBot are allowed on the pages you want cited. Then view the page source and confirm the key text is in the HTML, not loaded only by script.
4. Fix your ten most important pages. Put a direct answer in the first two sentences of each section. Add the specifics only you can give: prices or price ranges, timelines, limits, named standards, and what you do not do. Cite outside facts to their primary source.
5. Make your business facts consistent. Name, address, services, prices and founding details should match across your site, Google Business Profile, Bing Places, LinkedIn and key directories. Add Organization and relevant schema that matches the visible text on the page.
6. Re-test monthly and track trends. Run the same questions again every month. Note new citations, lost ones and wrong statements about you, and fix the pages behind them first.

Worked example: a 12-person accounting firm

Here is an illustrative example, not a client result. A firm tests 24 questions such as “how much does a small business tax return cost in Ohio”. It finds it is cited in 3 answers, while two larger firms are cited in 15. Their pages state fee ranges and deadlines; the firm’s pricing page says only “contact us”.

The firm rewrites its pricing, services and FAQ pages with fee ranges, inclusions, filing deadlines linked to the IRS and a short answer at the top of each, and fixes a firewall rule that blocked a crawler. At the next monthly test it compares citations per question, not total traffic, and keeps what moved a citation.

For keeping new articles to this standard, our AI content workflow that protects quality and brand voice shows how to brief, draft, fact-check and publish without filling your site with generic copy.

How do you measure visibility in AI search?

You measure AI search visibility with three sources: platform reports, your own prompt tests and conversions. Clicks alone will mislead you. A Pew Research Center study of 900 US adults found that people clicked a traditional result on 8% of Google visits when an AI summary appeared, against 15% when none did, and clicked a link inside the summary only 1% of the time.

• Google Search Console: Google points site owners to a Generative AI performance report for how content performs in its AI features.
• Bing Webmaster Tools: since February 2026, the AI Performance report shows total citations and average cited pages per day across Copilot, Bing AI summaries and some partner integrations.
• Your prompt log: the monthly spreadsheet from step 6, which is the only view across ChatGPT, Gemini and Perplexity together.
• Conversions: leads, sales and sign-ups from AI referrals in your analytics. Google reports that clicks from results pages with AI Overviews are higher quality, with visitors more likely to spend more time on the site.

How to read your prompt log

Each month, work out your citation rate: questions where any assistant cites one of your pages, divided by questions tested. Then match the pattern to its likely cause.

Pattern in your log	Likely cause	Fix first
Never cited, on any question	Crawler blocked or pages not indexed	robots.txt, CDN and firewall rules, then indexing in Search Console and Bing Webmaster Tools
Cited on brand questions, absent on “how much” and “X vs Y” questions	Pages lack specific facts	Add prices or ranges, timelines, limits and comparisons
Cited, but the answer misstates your facts	Conflicting details across your site and profiles	Make name, services and prices identical everywhere
Competitors cited from a directory or review site	The assistant trusts that third-party source	Get listed accurately there, by legitimate means

Then expect movement over months, not days. Nobody can promise a citation, and any service that guarantees one is overselling.

Common GEO mistakes to avoid

• Vague claims such as “fast turnaround” where a number would do, like “delivery in 10–20 working days”.
• Key facts inside images, PDFs, sliders or tabs that load only after a click.
• Buying placements or posting fake mentions on forums. It breaks platform rules and can backfire on your brand.

Accessible, well-structured pages help here too, because assistive technology and AI agents both read the structure of your page. Our guide to WCAG 2.2 fixes for business websites covers the same headings, labels and alt text that make pages easier for machines to read.

How Upmark Labs can help

If you want evidence before you change anything, start with our AI Search Visibility Audit, from $1,625. We turn your buyer questions into 25 test prompts, run each one in ChatGPT, Gemini and Perplexity, check every answer by hand, map the sources cited and count how often three named competitors appear. You get a report, a source map and a ranked fix list in 10–20 working days, walked through on a recorded call.

If you want the fixes made as well, the Search and AI Visibility Foundations Bundle, from $5,000, adds an SEO audit of up to 200 pages and answer-first rewrites of ten key pages with validated schema and a crawler check, in 20–30 working days. You provide Search Console access, editor access to your site, three competitors and one person to approve changes, within 72 hours of ordering. The scope and price are fixed on the Scope Card before you pay, and all work happens in your own accounts.

Neither package is right for everyone. If you want answers tracked every month rather than a one-off snapshot, the AI Search Visibility Monitor, at $675 a month with a three-month minimum, fits better. If your main problem is classic rankings or technical site health, start with our SEO Audit and Roadmap instead.

Frequently asked questions

A sensible next step is to measure before you rewrite. Order the AI Search Visibility Audit to see exactly what ChatGPT, Gemini and Perplexity say about you today, or book a free expert consultation if you would like to talk through your situation first.