Answer engine optimization

AEO: what six answer engines actually cite

Answer engine optimization (AEO) is the work of making a brand present and quotable in the single answer an AI assistant gives, rather than competing for a position on a page of links. It has two halves: being described on the sources the engine cites, and writing pages an engine can lift an answer from.

Most AEO advice is written from other AEO advice. This page is written from transcripts. We put real buyer questions to six engines with the web open and kept everything: 363 answers, 5,148 citations, every transcript published on this site.

Where a citation actually comes from

An exposed citation is the end of a three-step pipeline, and each step is a separate place to win or lose:

Retrieval. The engine searches, several times, with rewritten versions of the question. If your page cannot be found by ordinary search for any phrasing of the intent, nothing downstream can save it. This step also has a gatekeeper people forget: the retrieval crawlers themselves. Several engines fetch with agents that are not their training bots, and a blanket bot-blocking rule at the CDN quietly removes you from consideration.
Selection. Of the pages retrieved, a few are actually read into the answer. Pages shaped like the answer win here, which is why comparison pages, reviews and lists keep appearing in our citations: the engine is about to write a shortlist, and they already are one.
Attribution. The engine decides what to expose. This is where the engines diverge most sharply, and the table below is that divergence measured. Some show everything they touched; one major engine usually shows nothing at all.

The engines do not even agree on whether to show sources

Everything in AEO depends on a citation being visible: a source nobody can see cannot be verified, clicked, or earned. So the first number that matters is how many distinct sites each engine actually exposes per answer.

Engine	Answers	Show any source	Distinct sites cited
Grok	48	98%	36.8
Perplexity	63	100%	16.6
Claude	63	100%	6.6
ChatGPT	63	95%	6.3
DeepSeek	63	100%	2.9
Gemini	63	24%	0.2

Read the spread, because it is the strategy. The same question hands out roughly 36.8 visible citation slots on Grok and effectively none on Gemini, which shows a source in only 24% of its answers. On citation-rich engines you can earn your way into the sources. On citation-poor ones the answer comes from what the model already believes, which is built long before the question is asked, from how widely and accurately you are described.

Counted per answer as distinct cited domains, from our recorded transcripts. An engine may consult more than it exposes; exposed is what an AEO strategy can act on.

What actually shows up in the citations

The industry has a standard list: Reddit, Wikipedia, and you are done. Our citations tell it differently. Share of all recorded answers citing each destination at least once:

A review platform (G2, Capterra, Trustpilot...)10%
Reddit6%
An app store6%
YouTube3%
Wikipedia2%

Two things worth sitting with. Review platforms beat Reddit in our corpus, which is not what the advice industry says. And Wikipedia, the site everyone treats as the whole game, is nearly absent from visible citations: its influence lives in the memory layer, what the model already believes, rather than in live retrieval. That memory layer is real (14% of all answers cited nothing at all), but you reach it through years of broad description, not through one page.

The rest of the citations spread across 1,952 distinct domains, with the top 20 holding only 9%. The long tail is the game.

Writing the page that gets lifted

Retrieval gets a page seen. Extraction is what quotes it. The pages that keep showing up in our citations share a shape, and it is learnable:

Answer in the first forty words

The lifted passage is almost always a self-contained block near the top: a definition, a verdict, a number with its context. If your answer arrives in paragraph six, the engine quotes whoever put it in paragraph one. This page opens with exactly such a block, on purpose.

Make headings questions, and answer them immediately

Engines match a question to a passage. A heading that is the question, followed by a direct answer, is the easiest possible match. A clever heading is a miss.

Contain something that cannot be paraphrased away

A specific number, a dated measurement, a named method. Generic advice gets absorbed without credit; a figure gets quoted with its source. Original data is the only content with a built-in citation.

Be one thing per page

A page that answers one question completely beats a page that answers ten questions partially, because extraction happens at passage level and dilution costs matches.

Stay dated and maintained

Answer engines favour pages that look current, and readers who arrive from an answer bounce off stale ones. A visible updated date, kept honest, is cheap credibility on both fronts.

The shapes engines lift, with examples

Extraction is pattern matching, and the patterns are visible if you read enough answers. Four shapes keep getting quoted, and each has a template you can use today:

The definition block

A term, then what it is, in one self-contained breath. “X is the practice of..., used when...” The trap: burying the definition after three paragraphs of throat-clearing about how fast the industry is changing. The engine quotes whoever defined it cleanly, not whoever discussed it longest.

The verdict sentence

“For a two-person team, X is the better choice because Y.” Engines are asked for recommendations constantly, and a page willing to conclude gives them a conclusion to lift. Pages that end every comparison with “it depends on your needs” are unquotable by construction.

The number with its context

“Across N measured answers, X happened Y% of the time.” A figure travels with attribution attached, because restating it without the source makes the restater look unreliable. This is the only content shape with a built-in citation, and it is why every number on this page carries its denominator.

The comparison row

Two named things, one stated difference. Tables full of these get dismantled and quoted piecewise. It is also, not coincidentally, the shape of the shortlist the engine is about to write.

Anatomy of a cited answer

Here is what all of the above looks like in the wild: the most heavily sourced answer in our corpus for a high-volume question, reproduced exactly, with its citation count on it. Notice what kind of pages fill the slots.

A real recorded answer, not an illustration

“What are the best free AI video generators available?”

from “best free ai video generator” · 8,100 searches a month · 6 models answered

Grok91 sources · 71 distinct sites

Free AI video tools are usable in 2026, but none are unlimited, high-quality, watermark-free, and commercially licensed at the same time. Expect short clips (usually about 5–10 seconds), credit limits, and terms that change often. Confirm current limits on each site before you rely on them.

Best free options right now

Tool	Best for	Free access (typical)	Main limits
Google Veo (via Google Flow / Gemini)	Highest quality, with native audio	Daily Flow credits on a free Google account (often cited as ~50/day, enough for a few short clips)	~8-second clips; best models may be out of reach on free credits; SynthID metadata is common
Kling AI	Realistic motion and daily practice	Refreshing daily credits (widely reported around 66/day)	Short clips, usually watermarked, lower resolution, personal use only on free
Hailuo (MiniMax)	Character and expressive motion	Free or daily trial credits (amounts vary by region and promotion)	Short clips (~6s); watermark and commercial rights often restricted
CapCut	Finishing a real video for free	Full free editor, 1080p export, no watermark on standard timeline exports	AI generation itself is limited; this is an editor first
Canva	Template-based social clips	Free plan with limited AI video (Magic Media)	AI generations are capped; best for assembly, not cinematic scenes
Pika	Stylized effects and social experiments	Free account exists; monthly free credits have been reduced or removed in some recent pricing views	Lower resolution historically; commercial use usually paid
HeyGen	Talking-head / avatar videos	A few short videos per month	Watermarked; not open text-to-video
Open-source (Wan, LTX-Video, CogVideoX)	Unlimited local generation	Free if you run it yourself	Needs a capable GPU and setup time

How to choose

Best raw quality for free: Start with Google Veo in Google Flow or the Gemini app. It is the option most recent reviews single out for cinematic look and synchronized sound on a free tier.
Most sustainable free experimentation: Kling is the usual pick when you want several realistic clips every day without paying. Free exports are generally watermarked and not meant for commercial use.
Best free tool if you need something you can actually post: CapCut. Generate a short clip elsewhere, then edit, caption, and export at 1080p without a watermark on normal exports. Canva is the simpler template route.
Avatars and explainers: HeyGen (and, more limited, Synthesia) if you want a presenter reading a script, not a generated scene.
No credit cap at all: Run an open model such as Wan locally. Quality can be competitive, but you need hardware.

What to expect from “free”

Clips are short. Longer videos mean stitching several generations in an editor.
Runway still offers a useful one-time credit grant for new accounts, but those credits do not refresh, so it is a trial, not an ongoing free plan.
Visible watermarks, invisible identifiers (such as Google’s SynthID), and “personal use only” clauses are common on free tiers.
OpenAI’s consumer Sora app is no longer a free option; multiple 2026 reports say the standalone app was shut down.

Practical combo: generate B-roll in Google Flow or Kling, then assemble, caption, and export in CapCut. That covers quality and a clean finished file without a subscription.

Limits move quickly. Check each product’s pricing or account page before you plan a project around a specific credit number.

Sources the model used

What this answer is made of: 91 citations across 71 distinct sites, and the brands it names arrive with those sources attached. Every one of those citation slots is a page somebody could have been described on, which is the entire mechanism this guide is about.

Read the full transcript, every model, every source

And the half that happens off your site: the answer is assembled mostly from pages you do not own. Being accurately described on the roundups, reviews and comparisons an engine already cites in your category is the other, larger part of the work, and it is the part we do for companies.

The AEO checklist

Everything above, reduced to what you can verify page by page in an afternoon:

01The page answers its question inside the first forty words, in a block that stands alone.
02The main heading is the question a buyer would type, not a clever version of it.
03There is at least one thing on the page nobody can paraphrase away: a number, a dated measurement, a named method.
04Every comparison reaches a conclusion for a named situation instead of ending in "it depends".
05One page, one question. Adjacent questions get their own pages, linked.
06A visible, honest updated date, and content that justifies it.
07The retrieval crawlers can actually fetch the page: no blanket bot block at the CDN, no login wall on the content.
08The brand is described identically where it matters: same category words on your site, your profiles, and the third-party pages that mention you.
09You appear on at least some of the comparison and review pages the engines already cite in your category.
10You re-check the answers on a schedule, against the same questions, and keep the transcripts.

See whose pages the engines cite in your category

Run the free check: we ask your buyers’ real questions to the live engines and show you every answer with every exposed source, including whether any of them is yours.

Run the free check

Free, no account, no card.

FAQ

Frequently asked questions

What is answer engine optimization?▾

Answer engine optimization, or AEO, is the work of making a brand present and quotable in the one answer an AI assistant gives, instead of competing for a position on a page of links. In practice it means two things: being described on the sources the engine cites, and writing pages an engine can lift an answer from.

Is AEO different from GEO or AI SEO?▾

They are names for the same job with different emphasis. AEO stresses the output, one answer rather than ten links. GEO stresses the systems, engines that generate. AI SEO is the broadest label and also gets used for the separate topic of doing SEO work with AI tools. The work underneath is identical.

Do all answer engines cite their sources?▾

No, and the differences are larger than most advice assumes. In our recorded answers, the number of distinct sites an engine exposes per answer ranges from around 37 at the top to under one at the bottom, and one major engine shows any source at all in only about a quarter of its answers. A citation strategy has to be engine-aware.

Does being cited require backlinks?▾

No. Engines cite pages, not link graphs. A page gets cited because it is retrievable, matches the question, and contains something worth lifting: a clear claim, a number, a definition. Authority helps a page get retrieved, but the extraction step is about what is on the page.

What kind of page earns citations most easily?▾

Pages that answer a question in their first breath and contain something specific: original numbers, a clean definition, a direct comparison. In our corpus, review platforms and comparison pages appear in citations more often than brand-owned pages, which almost never get to vouch for their own brand.

Which answer engines matter most right now?▾

The one your buyers use, which you find out by asking them, and then the citation-rich ones, because those are the only engines where earning a visible source slot is even possible. In our corpus the depth ranges from about 37 exposed sites per answer down to effectively zero, and effort should follow that gradient.

Does AEO work for local businesses?▾

Yes, and often faster than for national brands. Local questions retrieve local pages, directories, community threads and review platforms, which are exactly the citation shapes an engine reaches for, and far less crowded than national comparison pages. The mechanism is identical; the competition is thinner.

How is AEO measured?▾

Record the answers, not a proxy. A fixed set of real buyer questions, asked on a schedule to each engine, keeping the full text and the exposed sources. Position in a list has no meaning here; presence, phrasing and citations do.

Can I just add FAQ schema and rank in answers?▾

Schema helps machines parse a page, and it is worth having. It does not manufacture a reason to cite you. An engine lifts substance, and no amount of markup makes an empty page substantial.