AEO and GEO Explained: How to Show Up in AI Answers

AEO (answer engine optimization) and GEO (generative engine optimization) are the practice of getting your brand named or cited when someone asks an AI assistant a question instead of scanning a list of links. AEO focuses on being picked as the single direct answer, whether that is a featured snippet, a voice reply or a one-line response from an assistant. GEO focuses on being one of the sources an AI engine draws on and credits when it writes a longer answer in ChatGPT, Perplexity, Gemini, Claude or Google AI Mode.

Both depend on two things. The first is whether your own pages are easy to retrieve and easy to quote. The second, and the one most brands underinvest in, is whether the rest of the web vouches for you: whether third-party articles, “best of” listicles, trade publications, press and community threads already mention your brand in the context the buyer is asking about. This guide explains how the engines work, which levers move the needle on-site and off-site, how to measure progress, and what a 90-day plan looks like.

AI engines name the brands that appear consistently in sources they already trust, so AEO and GEO are won mostly on other people's pages and finished on your own.

What are SEO, AEO and GEO, and how do they differ?

SEO (search engine optimization) is the work of getting pages to rank in a conventional results list and earning the click. Its levers are well known: keyword targeting, technical health, internal structure and backlinks from other sites.

AEO is a narrower target. The question has one answer, and the engine will show only one. Featured snippets, People Also Ask boxes, voice assistant replies and the short factual answers from AI chatbots all fall here. The content job is to state the answer plainly in a form a machine can lift: a definition, a number, a short list, a step sequence.

GEO is the newest of the three and the term comes from research rather than marketing. Researchers from Princeton and IIT Delhi published “GEO: Generative Engine Optimization” in November 2023, later presented at KDD 2024. They defined generative engines as systems that retrieve several sources and synthesize one response, and they built a 10,000-query benchmark to test which changes to a source page made it more visible in those responses. Their headline finding was that visibility could rise by up to 40 percent, and that the most reliable improvements came from adding statistics, quotations from credible sources and citations. Keyword stuffing, the old SEO reflex, offered little to no improvement and in some tests scored below the unoptimized baseline.

The three overlap more than the acronyms suggest. The table shows where they differ.

The practical reading is that SEO is the foundation, AEO is a formatting discipline on top of it, and GEO adds a third layer that is mostly about reputation across the web. A brand that is strong in SEO but invisible in AI answers almost always has a GEO gap, not an SEO one.

How do AI engines build an answer?

Understanding the pipeline removes most of the mystery. Every major AI search product, whatever the brand, runs a version of the same four steps.

Step one: the engine rewrites the question. A single prompt such as “best accounting software for a small manufacturing business” is broken into several related searches: accounting software for manufacturers, inventory accounting features, pricing comparisons, user reviews. Google calls this “query fan-out” and describes it in its own documentation on AI features and your website, noting that both AI Overviews and AI Mode may issue multiple related searches to build a response. Perplexity and ChatGPT search do the same thing with their own search backends.

Step two: the engine retrieves candidate pages. For each sub-query it pulls a set of pages from a search index. Google uses its own; ChatGPT has relied on Bing and its own crawler; Perplexity runs its own index. This is why SEO still matters. A page that does not rank for any of the fan-out queries is never in the room. Google’s documentation says the SEO best practices it already recommends are what apply to AI Overviews and AI Mode; there is no separate checklist to qualify.

Step three: the model reads the candidates and writes. The language model is given the retrieved passages and asked to answer. It favors passages that are self-contained, specific and consistent with each other. When several independent sources agree that a brand belongs in a category, the model treats that as settled and includes the brand. When only the brand’s own site says so, the model has one voice arguing its own case and often leaves it out.

Step four: the engine attaches citations. Each engine has a different habit here. Perplexity cites heavily, often a dozen or more sources per answer, and shows them inline. ChatGPT search cites fewer sources, which makes each citation worth more. Google AI Overviews show a panel of links drawn from the pages the answer was grounded in. In every case the cited URL is usually the page where the specific claim was found, which is often a third-party review or comparison rather than the brand’s homepage.

Two consequences follow. First, the engines reward pages that answer one sub-question cleanly, because fan-out means your page may be fetched for a question you never targeted. Second, the model is doing something closer to a literature review than a ranking. It is looking for agreement across sources. That is why the off-site half of GEO matters so much.

What decides whether a brand gets named?

The honest answer is that nobody outside the engines knows the exact weighting, and it shifts. But there is enough public evidence, and enough consistency in what we see across client prompt sets, to separate the levers that matter from the ones that are mostly folklore.

On-site levers you control directly

Quotable structure. The Princeton GEO paper tested nine content changes. The ones that lifted visibility most were adding statistics with sources, adding direct quotations from credible people, and adding citations to authoritative references. In the authors’ test on Perplexity, the best methods beat the unoptimized baseline by 22 to 37 percent depending on the metric. The lesson is not to pad pages with numbers; it is to make each page carry at least a few specific, attributable facts that a model can lift and credit.

Direct answers under question headings. Write the H2 as the question a buyer would type, then answer it in the first one or two sentences. Models extract passages, not pages. A 40-to-60-word paragraph that fully answers “how much does X cost” is more likely to be quoted than a 600-word section that gets there eventually.

FAQ sections. A short FAQ at the end of a page does two jobs. It gives the engine several more clean question-and-answer pairs to retrieve, and when marked up with FAQPage schema it is unambiguous to parse. Keep answers self-contained, 40 to 90 words, and avoid referring to “the section above”.

Schema and entity consistency. Schema (structured data in a standard vocabulary) tells machines what a page is: an article, an organization, a product, a FAQ. It matters less for the citation itself and more for entity resolution: helping the engine be sure that “Acme Polymers” on your site is the same “Acme Polymers” mentioned in a trade publication. Use the same brand name, founder name, address, product names and category descriptors on your site, your LinkedIn page, your Crunchbase or Companies House entry and your directory listings. Inconsistency makes you several weak entities instead of one strong one.

Freshness and crawlability. Engines that fetch live pages prefer recent ones. Keep a visible updated date, make sure your robots.txt does not block the AI crawlers you want (GPTBot, PerplexityBot, Google-Extended, ClaudeBot) and avoid loading the core answer through JavaScript that a crawler may not execute.

Off-site levers that carry the most weight

Third-party mentions. In May 2025 Ahrefs published an analysis of 75,000 brands and the factors that correlated with their visibility in Google AI Overviews. Branded web mentions came first at a Spearman correlation of 0.664. Branded anchor text was next at 0.527, branded search volume 0.392, Domain Rating 0.326, and the raw number of backlinks only 0.218. Roughly 26 percent of brands in the sample had no AI Overview mentions at all. Ahrefs is careful to say these are correlations, not causes. But the direction is clear: being talked about on other sites predicts AI visibility three times better than being linked to.

Profound’s analysis of 11.8 billion AI citation data points points the same way from the citation side. Brand-owned websites took about 57 percent of citations, so roughly 43 percent went to third-party sites such as media, communities and institutions, and on ChatGPT brand websites fell to 47 percent, meaning more than half of its citations went to third-party pages (Profound, 2026, summarized by WARC). Put the two studies together: the engines both see brands through third-party mentions and cite those third-party pages when they answer.

“Best of” listicles and comparison articles. When someone asks an assistant for “the best payroll providers for professional services firms” or “top flexible packaging suppliers in India”, the engine is retrieving exactly this kind of page and often reproducing its list. In our experience, getting into the listicles that already rank for your category is the single highest-leverage GEO move for most brands. It matters for a manufacturer, a law firm or a consumer app every bit as much as for a software company.

Trade press and digital PR. Coverage in a publication the engines already trust does three things: it is a mention, it is usually a link, and it is often the page that gets cited. Data-led stories (an original survey, a benchmark, an index) are the easiest to place because they give journalists a number to report, and they give the model exactly the kind of statistic the Princeton paper found most persuasive.

Community threads. Reddit, Quora, Stack Exchange and industry forums are among the most-cited sources in AI answers because they read as independent opinion. A thread where real users discuss your category and name your brand, honestly and in context, can be retrieved for dozens of prompt variants. Astroturfed threads get removed and can poison the brand’s entity, so this work has to follow community rules.

In engagements we track, brands we have placed in third-party articles, guest posts and listicles have seen AI citations rise by roughly 70 to 80 percent on the prompts where those articles are cited. We do not claim that lift across every prompt, and we would be suspicious of anyone who does. The pattern is consistent, though: when a trusted page names the brand, the engines that retrieve that page start naming it too.

Which levers are overrated?

Three things get recommended constantly and move little on their own.

Schema as a magic bullet. Structured data is hygiene. There is no public evidence that FAQPage or Organization markup, by itself, earns AI citations. Do it, then move on.

An llms.txt file. This proposed convention lists your important pages for AI crawlers. It is harmless and cheap, but none of the major engines has committed to using it, and we have not seen it change visibility in any prompt set we track.

Volume of on-site content. Publishing more pages does not increase agreement across the web about who you are. It increases the number of places you say it. The Ahrefs study found the number of pages on a site correlated with AI Overview visibility at only 0.17, half the correlation of Domain Rating and a quarter of brand mentions.

How do you measure AEO and GEO?

AI answers are not stable the way rankings are. The same prompt can return a different list on Tuesday than on Monday, and different users see different answers. Measurement therefore has to be sampled and repeated, not checked once.

Build a prompt set. Write 30 to 100 prompts that your buyers actually ask, in their words. Mix the types: category prompts (“best X for Y”), comparison prompts (“X vs Z”), problem prompts (“how do I fix…”), and branded prompts (“is Brand A any good”). Include industry variants if you sell to several, since a fintech buyer and a manufacturing buyer phrase the same need differently.

Run it on a schedule. Monthly is the minimum; fortnightly is better during an active program. Run each prompt on ChatGPT, Perplexity, Gemini, Claude and Google AI Mode. Record four things per prompt and engine: whether your brand was named, its position if the answer is a list, the sentiment of what was said, and the URLs cited.

Compute share of answers. The core metric is the percentage of prompts where you were named, by engine. Track two or three competitors on the same prompt set so you have a baseline. A brand named in 12 percent of prompts that rises to 30 percent over a quarter has moved; a brand that is “cited more” with no denominator has a feeling, not a metric.

Map citations to sources. Keep a running list of which third-party pages are cited for your prompts. This is the list that drives the off-site plan. When a page you have been placed in starts appearing as a citation, you can connect the placement to the visibility change.

Watch referral traffic separately. Assistants send some traffic, and analytics can segment it by referrer. It is a useful trailing indicator but a poor leading one, because the value of being named in an answer is often the brand recall rather than the click.

Our free LLM visibility checker gives a quick first read on your AI visibility, and we go deeper on method in how to measure your brand’s AI visibility.

What does a 90-day AEO and GEO plan look like?

The sequence below is what we would run for a mid-sized brand in any sector. The dates are for planning, not promises; off-site work is dependent on editors and publishing calendars.

Two notes on sequencing. First, do the on-site work before the off-site work, because when a third-party article sends a crawler to your site you want the landing page to confirm the claim in quotable form. Second, do not wait for perfect on-site pages before starting outreach. Editorial lead times are the long pole in the plan; a guest post takes two to four weeks from pitch to publication, and a press campaign four to eight.

What are the common mistakes?

Treating GEO as a content-volume problem. The brands that struggle most are often the ones with the most blog posts. Volume on your own domain does not create agreement across the web. One placement in a cited listicle beats twenty new articles.

Chasing every engine with one tactic. Perplexity cites widely and favors fresh, extractable pages. ChatGPT cites few sources and leans on established media and communities. Google AI Overviews stay close to what already ranks. A plan that only publishes on-site content will do reasonably on Google and poorly on ChatGPT. Measure per engine and allocate accordingly.

Buying mentions on sites nobody trusts. A mention on a link farm does not count as independent agreement; at best it is ignored, at worst it associates your entity with spam. Vet every site: real editors, steady organic traffic, topical fit, and whether the engines already cite it. We cover our full process in how to vet a website before buying a link.

Inconsistent entity details. Different brand names across your site, LinkedIn and press (the legal name in one place, a trading name in another, an old product name in a third) split your authority. Pick one form of each and use it everywhere.

Faking community activity. Posting your own praise on Reddit from fresh accounts usually gets removed quickly and can get the domain filtered. Contribute where you have something genuine to add, disclose affiliation when the community requires it, and let placement survive on merit.

Measuring once. A single screenshot of an AI answer proves nothing. Build the prompt set, run it repeatedly, and report share of answers with a competitor baseline.

How does Model Marketing help?

Most of this guide can be done in-house. Where brands bring us in is the off-site half, because it needs relationships with editors, vetted publication inventory and the patience to place a brand into articles that already rank and already get cited. Over eight years we have placed more than 12,237 links and reached 10,132 publications, with 22 link builders and 9 full-time writers, for clients in SaaS, fintech, professional services, consumer apps, B2B services and manufacturing.

Our AEO and GEO service begins with a two-week audit: a prompt set across the five main engines, a baseline share of answers, an entity and on-site review, and a target list of the third-party pages being cited for your category. Scope and price follow the audit, and the plan combines on-site restructuring with placements through our link building, digital PR and Reddit marketing teams. Every site we place on passes nine checks, the ninth being how often it is already cited by ChatGPT, Perplexity, Gemini, Claude and Google AI Mode.

There is no retainer. Guest posts, insertions, Reddit placements and PR campaigns are billed per live unit, twice a month, with 30 days’ notice to stop and a 12-month free replacement of any link that is removed. Published rates are on the pricing page. In our experience AI visibility typically starts to move within 30 to 60 days of the first placements going live.