By
Austin Coker
Reading time: 15 min
In the research that named this field, the same content edits raised a fifth-ranked page’s visibility inside AI answers by 115% and lowered the top-ranked page’s by 30%. What decided the outcome wasn’t the edit. It was where the page already sat in the underlying results.
Generative engine optimization (GEO) is the practice of getting your content retrieved, quoted and cited by AI search engines like ChatGPT, Perplexity, Gemini and Google’s AI Overviews, instead of only competing for blue links. That ranking-position finding is the most useful thing a marketing leader can know before approving a GEO budget.
This guide covers what generative engine optimization is, how each engine retrieves sources, which methods have a measurable effect and which don’t, how GEO relates to SEO in 2026, how to measure it, and what the work looks like in different industries.
Generative engine optimization is the set of practices that increase how often, how prominently and how favorably your content appears in answers written by AI search engines.
The term and the first way to measure it come from the GEO paper by Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, researchers at IIT Delhi and Princeton, presented at KDD 2024. The authors built GEO-bench, a benchmark of queries across many subject areas, and tested nine content methods against it: authoritative tone, adding statistics, keyword stuffing, citing sources, adding quotations, easier language, fluency, unique words and technical terms.
Their main metric is position-adjusted word count: how much of your content the generated answer uses, and how early. A mention buried in the last sentence of a long answer isn’t worth the same as a sentence used in the opening paragraph, and the metric reflects that.
Two things follow. First, GEO is a content and retrieval problem, not a technical-file problem. Second, the unit of success is a citation inside an answer, not a position on a results page.
Every major AI search surface runs the same basic loop. It interprets the question, searches an index, retrieves a set of candidate pages, and writes an answer grounded in what it found. Your job is to be in that retrieved set, and then to be the easiest source in it to quote accurately.
A user asks one question. The engine runs many.
Google’s documentation on AI features and your website (updated December 10, 2025) says AI Overviews and AI Mode may use “query fan-out,” running several related searches across subtopics and data sources before writing a response. One question about choosing a vendor can expand into separate searches for pricing, integrations, alternatives, implementation time and compliance.
That changes what a page needs to contain. A page that answers only the headline question competes for one branch of the fan-out. A page that clearly answers several related subtopics can be retrieved for more of them. We cover the mechanics in our guide to query fan-out in AI search.
Google says there are no additional requirements to appear in AI Overviews or AI Mode. Your page needs to be indexed and eligible to show a snippet in Search.
That makes snippet controls the real risk. A nosnippet directive, a data-nosnippet attribute on an important section, a tight max-snippet value or a stray noindex can keep a strong page out of AI features. These settings often get added during a redesign or a migration and are never reviewed again.
On the OpenAI side, the Publishers and Developers FAQ states that “any public website can appear in ChatGPT search.” No partnership or feed is required.
These three get confused constantly, and the confusion costs visibility.
That’s why the first thing we check on every GEO engagement is robots.txt, CDN and firewall rules. A blanket AI-crawler block added in 2024 can quietly keep an otherwise strong site out of ChatGPT today.
The GEO paper tested each method on its own against a baseline position-adjusted word count of 19.3. Four methods produced meaningful lift.
Adding quotations scored 27.2, about 41% above baseline, the strongest single method tested. Relevant quotations from credible sources made AI answers use the content more, and earlier.
In practice, your pages should carry attributable statements from named people: analyst commentary, customer language from a recorded interview, a line from one of your own practitioners with a title attached. A quotation is a safe unit for an AI engine to reproduce.
Adding statistics scored 25.2, about 31% above baseline. Pages with concrete numbers got used more than pages describing the same thing in adjectives.
Two caveats. The number needs a visible source, or it reads like an unsupported claim. And it should be a number a buyer would actually want in an answer, not a vanity metric about your company.
Citing sources scored 24.6, about 27% above baseline. Linking to the material behind your claims increased how much of the page was used.
This cuts against years of SEO teams hoarding links. In GEO, visible sourcing signals that a page is safe to build an answer on.
Fluency optimization scored 24.7, about 28% above baseline. Clear, well-built prose got used more than dense or awkward prose covering the same ground.
It’s the least glamorous finding and probably the most actionable for B2B sites. A lot of enterprise content can’t be quoted because no single sentence in it stands on its own.
The authors also tested on Perplexity, a live AI search engine, and reported visibility gains of up to about 37%, with quotations improving position-adjusted word count by roughly 21%. The effect outside the benchmark is smaller, but it’s real.
This finding should shape where you spend.
In the paper’s ranking analysis, the best methods helped lower-ranked sources far more than top-ranked ones, and the top source actually lost visibility:
Source: Aggarwal et al., KDD 2024
GEO is a challenger’s tool. If you sit fourth or fifth for the queries that matter, this work can roughly double your presence in the answer. If you already rank first, don’t expect the same lift. Protect your crawl access and snippet eligibility, keep your sourcing strong, and put most of the effort into subtopics you don’t own yet.
The best method also varied by subject area:
For B2B teams, the takeaway is that a fintech compliance page and a SaaS comparison page shouldn’t get the same treatment. Regulated, factual topics reward citation density. Business topics reward clear writing. Advisory and opinion content rewards hard numbers.
We’ll run your buyers’ questions across ChatGPT, Perplexity, Gemini and Google AI Overviews and show you where you’re cited and where competitors show up instead.
On May 15, 2026, Google published Optimizing your website for generative AI features on Google Search (updated July 10, 2026). It’s the most direct statement any engine has made about wasted effort, and it lines up closely with the research.
Keyword stuffing was the only method in the GEO paper that did worse than doing nothing. It scored 17.7 against the 19.3 baseline, about 8% lower.
Google is blunt: “You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search.”
There’s a small industry around llms.txt right now. If you’ve already published one, leaving it up costs nothing. Building a process to maintain it takes time better spent on content.
Two more direct statements from Google: “Structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add,” and “There’s no requirement to break your content into tiny pieces for AI to better understand it.”
Structured data is still worth implementing for rich results in regular search, just not as a GEO tactic. Google also says you don’t need to write in a special way for AI search.
Google says “seeking inauthentic ‘mentions’ across the web isn’t as helpful as it might seem.”
This is easy to over-read. Third-party coverage matters, and our own client results depended on it. What doesn’t work is mass-placing your brand name in low-quality roundups built to be scraped. What does work is earning real coverage, with substance, on sites the engines already pull from for your category.
What Google recommends instead is “unique, expert-led content that provides value beyond common knowledge,” grounded in first-hand experience. The guide also points to AI agents and the Universal Commerce Protocol as an emerging area, which matters most for ecommerce.
Google’s position is clear: “optimizing for generative AI search is optimizing for the search experience, and thus still SEO.”
That’s a fair description of Google’s own products. AI Overviews and AI Mode draw on the same index, follow the same snippet controls and reward the same quality signals. There’s no second door into Google.
Everywhere else, it’s more complicated. ChatGPT runs its own crawler and search layer. Perplexity builds answers from its own retrieval. Microsoft’s Copilot and Bing AI summaries have their own citation behavior and reporting. A site can be excellently optimized for Google and still invisible in ChatGPT because of one robots.txt line, and standard SEO audits won’t catch it.
The other difference is the click. Pew Research found (July 22, 2025) that people clicked a regular search result on 8% of visits when an AI summary appeared, versus 15% when none did, and only about 1% clicked a link inside the summary. In its March 2025 sample, 18% of searches produced a summary, and Wikipedia, YouTube and Reddit were the most cited sources.
Google, for its part, says clicks from pages with AI Overviews tend to be higher quality, with visitors more likely to spend time on the site. Both can be true: fewer clicks, better clicks. We look at what that does to your funnel in our post on zero-click search.
So: on Google, GEO is a specialty within SEO, focused on being retrieved and quoted rather than just ranked. Across all AI search engines, it’s a distinct discipline with its own crawlers, measurement and failure points.
Attribution in AI search is harder than in organic search, and no dashboard fully solves it yet. Here’s what’s actually available as of October 2026.
On June 3, 2026, Google announced Search Generative AI performance reports in Search Console. They show impressions by page, country and device for AI features in Search, including AI Overviews and AI Mode, plus Discover. They’re rolling out to a subset of sites, with more metrics promised.
Before that, AI feature traffic was folded into the standard Performance report under the “Web” search type with no way to separate it. If you don’t have the new reports yet, treat that number as blended, and say so in reporting.
Bing’s reporting goes further. On February 10, 2026, Microsoft launched AI Performance in Bing Webmaster Tools in public preview, showing how often your pages are cited in Microsoft Copilot and Bing’s AI summaries, which pages, and the “grounding queries” behind them. Bing called it “an early step toward Generative Engine Optimization (GEO) tooling.”
On June 16, 2026, it added Intents, Topics, Citation Share and Compare. Citation Share is your citations as a percentage of all citations for a grounding query, the closest thing to share of voice in AI search, and it’s free. Even if Bing is a small channel for you, it’s a useful read on how retrievable your content is.
ChatGPT adds utm_source=chatgpt.com to referral links, per OpenAI’s Publishers and Developers FAQ. That gives you a clean analytics segment for ChatGPT visits that you can tie to conversions.
Judge that segment by what visitors do, not just how many there are. AI search sends fewer sessions than organic search, so raw volume will understate its value.
Put the Pew finding next to Google’s point about click quality and you get a practical approach: stop reporting AI search on impressions and sessions alone. Report on whether you’re cited, and on conversions from the visits you do get.
The question to answer isn’t how much traffic came from AI search. It’s whether your brand shows up when a buyer asks the question that comes right before a purchase, and what those people do next.
The figures below are 95 Projects’ own reported client results, not third-party verified data.
Two patterns stand out. Both combined on-site answer pages with coverage on sites the engines already pull from for that category, rather than relying on either alone. And both targeted questions with buying intent behind them, which is why the recruitment engagement produced revenue we could trace, not just impressions.
We build the pages and earn the third-party coverage that gets B2B, SaaS and ecommerce brands recommended in AI search.
The sources AI search engines lean on change by industry. These guides go into the specifics.
GEO for SaaS. Answers in software categories tend to draw on comparison and alternatives pages, clear pricing and product documentation. Documentation is the underused asset because it’s precise and easy to quote. For a ChatGPT-specific playbook, see our B2B SaaS ChatGPT guide.
GEO for recruitment agencies. Role- and sector-specific pages, salary and market context, and coverage in HR and careers publications tend to get cited. Generic agency homepages rarely do.
GEO for staffing agencies. Similar to recruitment, with more weight on the locations you cover, compliance credentials and named industry specialties.
GEO for IT services and MSPs. Certifications, vendor partner status, clear service detail and local market pages give AI search something concrete to cite.
GEO for fintech. Regulated topics reward dense, explicit sourcing, which matches the research finding for law and government content. Licensing and compliance detail belongs on crawlable pages.
GEO for ecommerce. Product attributes, reviews and third-party coverage carry most of the weight. With Google pointing to AI agents and commerce protocols, clean, machine-readable product data is becoming a priority for high-AOV catalogs.
GEO for manufacturing. Spec sheets, tolerances, materials, capabilities and certifications are what buyers ask about. Too often that information lives in sales decks and downloads instead of clear, crawlable pages.
GEO for cybersecurity SaaS. Original threat research, framework alignment and technical documentation give engines something credible to cite. Original research is the strongest asset here.
GEO for wealth management. Advisor credentials, clear fee structures, disclosures and minimums. AI engines are cautious with financial topics and favor sources that state qualifications plainly.
GEO for management consulting. Published research, named partner expertise, practice-area pages and methodology. Quotation-rich content fits naturally here.
Our GEO service follows a consistent sequence.
utm_source=chatgpt.com segment, and regular prompt testing against the baseline. We report on whether you’re cited and what those visits do, not on impressions alone.
Austin Coker is the CEO and founder of 95 Projects. He has worked in SEO since 2020, when he built and sold an ecommerce brand that grew mainly through search. He founded 95 Projects in 2022 to help SaaS, ecommerce, and B2B service companies show up wherever their buyers search, from Google to ChatGPT and AI Overviews.
Have questions about working with us? Book a 30-minute strategy call to discuss your goals and see if we’re a good fit.
No-pressure conversation
Clear next steps (if we’re a fit)
Talk directly to an expert
Generative engine optimization is the practice of making your content more likely to be retrieved, quoted and cited in answers from AI search engines like ChatGPT, Perplexity, Gemini and Google’s AI Overviews. The term comes from a KDD 2024 paper by researchers at IIT Delhi and Princeton, which introduced both the concept and a benchmark for measuring it.
On Google, not really. Google says optimizing for generative AI search is still SEO, because AI Overviews and AI Mode use the same index and snippet rules. Elsewhere it is different: ChatGPT, Perplexity and Bing run their own crawlers, retrieval and reporting, which traditional SEO work doesn’t cover.
Often faster than traditional SEO. In one of our ecommerce engagements, a brand went from absent to ChatGPT’s top recommendation for its target prompt in three weeks; a recruitment client saw $70,000+ in ChatGPT-referred business within four months. Competitive B2B categories take longer, so plan on three to six months for meaningful citation share.
No. Google says you don’t need new machine-readable files, AI text files, markup or Markdown to appear in Google Search, and Google Search ignores them. If you already have one, it does no harm. Just don’t build a maintenance process around it at the expense of content work that does change retrieval.
Not as a GEO tactic on Google. Google says structured data isn’t required for generative AI search and there’s no special schema.org markup to add. It’s still valuable for rich results in regular search, so keep using it for that reason.
ChatGPT adds utm_source=chatgpt.com to links it sends, so you can build a dedicated segment in Google Analytics or another analytics tool, as long as OAI-SearchBot isn’t blocked. Judge that segment on conversions rather than volume, and pair it with Bing’s Citation Share to see visibility that never produces a click.
Only if you don’t want your content used for model training. GPTBot is OpenAI’s training crawler. OAI-SearchBot is the one behind ChatGPT search, and blocking it keeps you out of ChatGPT’s answers. Mixing up the two is a common reason strong sites are missing from ChatGPT recommendations.
Often it works better. In the GEO research, citing sources lifted a fifth-ranked source’s visibility in AI answers by 115.1% while the top-ranked source dropped 30.3%. Adding quotations and statistics showed the same pattern. GEO favors challengers more than incumbents.
It depends on the industry. In SaaS, comparison pages and documentation; in regulated sectors, compliance and credentials content; in manufacturing, specs and certifications. Across all of them, pages with sourced statistics and attributable quotations performed best, with quotations scoring about 41% above baseline in the research.