Generative engine optimization, explained without the hype
Ranking well does not make a page citable. What an answer engine needs instead, and the page changes that make a claim liftable.
Generative engine optimization is the work of getting AI answers to cite you as their source. It overlaps with SEO, it is measured differently, and most sites are doing none of it.
What the term actually covers
Generative engine optimization (GEO) is optimising for systems that answer a question directly instead of returning ten links: ChatGPT, Claude, Perplexity, Gemini, Google’s AI Overviews. The output is a paragraph with two or three sources attached. You are either one of them or you are invisible.
That last part is what makes it a separate discipline. Ranking is graded on a curve across ten positions. Citation is close to binary per answer, and the number of slots is small.
Why ranking well does not carry over
Ranking well does not make a page citable. Position does not transfer, because an engine composing an answer prefers whichever source it can lift a clean claim from rather than whichever sits at the top. Click-through optimisation matters less when there is no click, and rendering tolerance is lower than Googlebot’s.
Position does not transfer. An engine composing an answer is not obliged to prefer the top result. It prefers the source it can most easily extract a clean claim from.
Click-through optimisation is beside the point. There is often no click. Your title tag and meta description, which exist to win a click from a results page, do less work when the result is a paragraph that quotes you.
Rendering tolerance is lower. Googlebot will come back and render your JavaScript on a second pass. Most answer-engine crawlers read the HTML they are given and move on.
What an engine needs from a page
An answer engine needs four things before it can cite a page: the content present in the HTML rather than painted in afterwards, a claim stated plainly enough to lift as written, something to attribute it to such as an author and a date, and permission to crawl the page at all.
- The content is in the HTML. Not painted in after hydration.
- The claim is extractable. Stated plainly, near a heading that resembles the question.
- There is something to attribute. A named author, a date, an organisation.
- The crawler is allowed in. Many
robots.txtfiles predate these agents and block them by omission.
Extractability, shown properly
Extractability is whether one passage on the page states the answer outright. It moves more than anything else on this list and it is a writing change rather than a technical one, because an engine quoting you needs a sentence it can take as written, without stitching it together from three paragraphs.
Before
We are often asked about the cost of a technical audit, and the honest answer is that it varies quite a lot depending on scope, the size of the site, and whether remediation is included, though most agencies land somewhere in a fairly broad band.
After
A technical SEO audit typically costs between $500 and $5,000. The range depends on three things: the number of pages crawled, whether remediation is included, and whether the audit is a one-off or part of a retainer.
The second version can be lifted as an answer. The first cannot, because there is no sentence in it that states the answer. Same information, same honesty, different shape.
Attribution, which is cheap and often missing
An engine deciding between two equally useful pages will often prefer the one it can characterise. Give it something:
<article>
<h1>How answer engines choose sources</h1>
<p class="byline">By Dana Whitfield, Head of Research at Corvid Analytics</p>
<time datetime="2026-03-11">11 March 2026</time>
</article>That is three lines. It costs nothing and it is the difference between an anonymous page and an attributable one.
Letting the right crawlers in
Check your robots.txt for the agents that matter now, not the ones that mattered in 2019:
User-agent: OAI-SearchBot
Allow: /
User-agent: PerplexityBot
Allow: /
User-agent: ClaudeBot
Allow: /Blocking them is a legitimate choice. Blocking them without knowing is not a choice at all.
How to measure whether any of it worked
Rank tracking cannot answer this, because the surface it measures is not the surface you are being cited on. What you want is citation share: for the queries you compete on, how often each engine names you, tracked over time.
A serviceable manual version costs an afternoon. Pick twenty questions your best pages answer, ask each of four engines, record who got cited, repeat monthly. That is a real baseline, and it will tell you more than a rank tracker will about this particular problem.
One caution on interpretation. Citation data is recorded per prompt and per engine, so it supports statements like “our share on this cluster of questions rose.” It does not support “posts with a summary box get cited 40% more,” because nothing links a citation back to the specific change that earned it. Be careful of anyone selling you that number.
What to do this week
Pick your five best pages. For each: check the raw HTML contains the content, add a one-sentence direct answer under the main heading, add author and date, and confirm the answer-engine crawlers are permitted. That is a couple of hours and it covers the four properties above.
The free Crawld scan runs 16 answer-engine checks as part of its rubric and reports which ones it could not measure. The GEO page covers what each check looks for, and if you want the underlying distinction spelled out, crawled, indexed and cited are three different states.