A growing share of the questions your customers used to type into Google are now typed into an assistant instead, and answered in a paragraph that may or may not mention you. Generative engine optimisation — GEO, also called answer engine optimisation — is the practice of making your content the material those systems draw on and cite.
There is a great deal of noise around this, much of it from people selling a product. This is a practical account of what actually appears to influence citation, what carries over from conventional SEO, what is genuinely new, and what to measure.
What Changes, and What Does Not
Start with the reassuring part: the foundations are the same. Generative systems are largely drawing on the same crawled web, and in many cases on live search results, so a page that cannot be crawled, loads slowly or has no authority behind it is invisible to both.
What changes is the unit of competition. Traditional SEO competes for a position; the page is the unit and the click is the outcome. Generative search competes for inclusion in an answer; the passage is the unit and a citation is the outcome. That single shift explains almost every tactical difference that follows.
The second change is that the click may never come. A user who gets a satisfactory answer with your brand named in it has been influenced by you without visiting you. That is real value, and it is invisible in a session-count report — which is why measurement has to change too.
Write in Extractable Passages
This is the highest-leverage change to how content is written, and it is a structural change rather than a stylistic one.
A generative system assembling an answer needs a self-contained chunk of text that answers a specific question without requiring the surrounding page for context. Content built as a flowing narrative — where paragraph four only makes sense after paragraphs one to three — is difficult to extract from. Content built as a series of complete, self-standing answers is easy.
In practice that means:
- Headings phrased as the question a person would ask, in their words rather than industry vocabulary.
- The direct answer in the first sentence or two after the heading, followed by the explanation. Not the other way round.
- Named entities restated rather than pronouned. “Spravato is administered under supervision” is extractable; “It is administered under supervision” is not.
- Specifics over generalities. Numbers, timeframes, thresholds and named standards are cited far more readily than adjectives, because they are checkable.
- Lists and tables for enumerable content, which chunk cleanly.
The useful mental test is to take any single section of your page out of context and ask whether it still answers something on its own. If it does not, it will not be quoted.
Be the Source, Not a Summary of One
Generative systems are synthesising from many documents. A page that summarises what everyone else already says adds nothing to that synthesis and has no reason to be cited. A page containing something available nowhere else does.
The kinds of material that earn citation:
- Original data. Your own numbers — pricing ranges you actually quote, timelines you actually observe, failure rates you actually see. Nobody else has them.
- Operational specifics. How a process really runs, including the parts that go wrong. This is the material generic content omits because the writer has never done the work.
- Explicit limitations. Saying when your solution is the wrong answer is unusual enough that it stands out, and it is exactly what a balanced synthesis needs.
- Local and jurisdictional detail. Regulation that varies by state or city, and what it means practically.
- Direct comparison. Structured comparisons against alternatives are heavily drawn on for “X vs Y” questions.
The corollary is uncomfortable but worth stating: thin content is now worse than useless. It was previously a weak asset that might rank for something. It is now a page with no reason to be included in any answer at all.
Entity Clarity and Consistency
These systems reason about entities — organisations, people, products, places — and the confidence with which they associate your brand with a topic depends on how consistently that association appears across the web.
- Describe your organisation identically everywhere — site, profiles, directories, press. Inconsistent descriptions produce a fuzzy entity.
- Structured data for Organization, Person, Product, Service and FAQ where genuinely applicable. It states the relationships explicitly rather than leaving them to be inferred.
- Real author entities. Named people with credentials, consistent across your site and their professional profiles.
- Third-party corroboration. What others say about you carries more weight than what you say about yourself — industry directories, associations, press, review platforms.
A practice that names its clinicians, states their credentials, and has those credentials corroborated on association registries is a far more confident entity than one whose site says “our experienced team”.
Technical Access
Some unglamorous items decide whether any of the above is even visible.
- Content must be in the server-rendered HTML. Text that only appears after client-side JavaScript executes is unreliable at best. This alone excludes a surprising number of modern sites.
- Decide your crawler policy deliberately. AI crawlers can be allowed or disallowed in robots.txt, and several are distinct from the traditional search crawlers. Blocking them is a legitimate choice; making it by accident is not.
- Content behind interaction is invisible. Accordions, tabs and modals that load on click hide their content. If it matters, render it.
- Speed and stability still gate crawling and indexing. See our overview of technical SEO.
GEO and Conventional SEO Compared
Setting the two side by side makes the re-weighting concrete rather than abstract.
- Unit of competition. SEO competes page against page for a ranking position. GEO competes passage against passage for inclusion in a synthesised answer.
- Definition of success. SEO: a click. GEO: a citation, which may produce no click at all and still influence a decision.
- Title and heading strategy. SEO rewards a compelling title that earns the click. GEO rewards a heading that exactly matches the question, because the match is what surfaces the passage.
- Length. SEO tolerates and sometimes rewards comprehensive long pages. GEO rewards well-structured pages of any length, and heavily penalises material where the answer is buried three hundred words in.
- Keyword usage. SEO thinks in terms of target phrases. GEO thinks in terms of entities and the relationships between them, so restating names is more useful than repeating a phrase.
- Freshness. Both value it, but generative systems that consult live search are noticeably sensitive to whether a page states a date and whether its claims are current.
- Authority. SEO weighs links heavily. GEO appears to weigh corroboration across independent sources — being described consistently by others, not only linked by them.
Note how few of those are in conflict. Almost every GEO practice also improves conventional performance, which is the main reason to treat this as an adjustment to your existing programme rather than a separate budget line.
Common Mistakes
- Stuffing pages with FAQ schema that does not reflect real questions. Markup describes content; it does not substitute for it.
- Publishing AI-written summaries of existing material. This is the exact category of content that has no reason to be cited, produced at scale.
- Blocking AI crawlers unintentionally through an inherited robots.txt or a security plugin’s default rules, then wondering why you never appear.
- Treating it as a separate content stream. A “GEO page” alongside an “SEO page” on the same subject is just cannibalisation with a new name.
- Reporting on it with session counts, which measure the one thing this channel is least likely to produce.
Measuring GEO
Conventional analytics will not show you this, and pretending otherwise is where most GEO reporting falls down.
- Prompt testing. Maintain a list of the questions your customers actually ask, run them across the major assistants on a schedule, and record whether you are cited and how you are characterised. Manual, unglamorous, and the only direct measure.
- Referral traffic from assistant domains in your analytics. Volumes are small; the trend is what matters.
- Branded search volume. If assistants are naming you, more people search your name. This is frequently the first measurable effect.
- Direct traffic and enquiry quality. Enquiries that arrive already well-informed, from people who cannot say how they found you, are a recognisable signature.
- Share of citation against named competitors across your prompt set — the closest equivalent to share of voice.
Be wary of anyone selling guaranteed AI visibility. These systems are non-deterministic, they change without notice, and the same prompt can return different sources on consecutive runs. Directional improvement over time is a reasonable goal. A guaranteed placement is not a thing.
Where This Leaves Your Strategy
GEO is not a replacement discipline and it does not require abandoning what works. The practical position is that it is a re-weighting: the things that were already good practice — genuine expertise, clear structure, specific claims, real authorship, clean technical delivery — now matter considerably more, and the things that were already marginal now return nothing at all.
If you want a single change to make first, it is this: go through your most important pages and rewrite each heading as the question a customer would ask, then make sure the first two sentences underneath answer it completely. It improves conventional rankings, it improves the experience for readers who skim, and it is the format these systems draw from most readily. Our note on how AI is transforming marketing strategies covers the wider context, and SEO for medical practices shows what the structural discipline looks like applied to one regulated vertical.
The businesses that will do well here are the ones that actually know something and are willing to write it down precisely. That has always been true. It is now simply harder to fake.
