What actually makes a page citable
Assistants quote passages, not pages. The unit that gets cited is a paragraph that answers a specific question completely, without needing the surrounding article for context. A page organised as a narrative — building an argument over eight hundred words before the payoff — is excellent for a human reader and nearly useless as a citation source.
This does not mean writing for machines. It means front-loading: state the answer, then explain it. A page that opens with its conclusion serves the impatient human and the extraction model equally well, and there is no page we have written where doing this made it worse to read.
The consistency problem nobody budgets for
Assistants weight information that agrees with itself across independent sources. If your site says one thing, your LinkedIn profile says another, and three directory listings each carry a different service description, none of it is confidently citable — the model has no basis for choosing between them.
This is why the highest-leverage GEO work is often not on your website at all. It is making the twenty places that already describe your company agree with each other: directory profiles, review sites, partner pages, social bios, old press mentions. Unglamorous work, rarely proposed by agencies because it is hard to bill as a recurring retainer, and consistently effective.
How to measure work that produces no clicks
This is the reporting problem that causes good programmes to be cancelled. If a buyer asks an assistant for recommendations, reads a summary that includes you, and then searches your name directly, your analytics records a branded direct visit. The work that created the demand is invisible to the tool measuring it.
- Track branded search volume as a primary metric. Assistant mentions produce searches for your name, and that is where the effect surfaces first.
- Query the assistants directly, on a schedule. Ask ChatGPT, Claude, Perplexity and Google's AI surfaces the questions your buyers ask, monthly, and record whether you appear and what is said about you.
- Watch impressions and average position in Search Console independently of clicks. Rising impressions against flat clicks is evidence of answer-surface visibility, not failure.
- Instrument the first conversation. Ask every new enquiry how they found you and store it as a field, not as anecdote.
- Accept a longer feedback loop. This compounds across quarters; judged month to month it will always look like it is not working.
What we would do in the first ninety days
Fix the technical floor first — rendering, speed, structured data, crawler access — because nothing downstream compensates for a page an engine cannot read or will not wait for. Then rebuild the pages that answer real buying questions so each leads with its answer rather than arriving at one. Then reconcile the off-site footprint so the web agrees about who you are and what you do.
Only after those three is content volume worth funding. Publishing into a site with an unresolved technical floor and an inconsistent footprint is the most reliable way to spend a year producing work that never compounds.
How do SEO, AEO and GEO fit together?
They get sold as three services because three retainers bill better than one. The groundwork is largely shared: a page a search engine can crawl, parse and rank is most of the way to a page an answer engine can quote, and both depend on the same structured data and the same clarity about what the page actually asserts.
Where they genuinely differ is the unit that wins. Classic SEO competes at the page level for a ranked position. Answer-engine optimisation competes at the passage level — a self-contained block that can be lifted whole and still make sense. Generative-engine optimisation competes at the entity level, on whether the wider web corroborates who you are consistently enough for a model to state it confidently.
Buying them separately pays three times for the shared foundation and produces three sets of recommendations that contradict each other on the specifics.
| Technical foundation | Rendering, speed, crawler access, structured data, clean information architecture. Serves all three outputs and is the prerequisite for every one of them. |
|---|---|
| SEO — the page | Competes for a ranked position on a query. Depends on relevance, authority and the technical floor being sound. |
| AEO — the passage | Competes to be the block an assistant quotes. Depends on self-contained answers, comparison tables, and answering the question in the first sixty words. |
| GEO — the entity | Competes on whether the web agrees about who you are. Depends on consistent descriptions across every profile, directory and mention that already exists. |
What a citable page actually looks like
There is a measurable shape to the passages assistants quote, and most B2B pages miss it in the same two ways. The answer arrives too late — after four paragraphs of context-setting that a human skims and a model discards — and the passage is not self-contained, so lifting it out of the page leaves it meaningless.
The fix is structural rather than stylistic. Answer the question the page is named after in the first sixty words. Make each answer block long enough to stand alone but short enough to quote whole; passages in the 134-to-167-word band get cited noticeably more often than shorter or longer ones. Put comparative information in a real table, because engines parse tables as structured data and cite them far more readily than the same facts written as prose.
None of that makes a page worse to read. A page that opens with its conclusion serves the impatient human and the extraction model equally well, which is why we have never had to trade one against the other.
How to evaluate anyone selling AI-search visibility
- Ask what they will do off your website. If the whole plan is on-site content, they are selling SEO with new vocabulary — entity consistency is off-site work by definition.
- Ask how they will measure it, specifically. "Assistant mentions" is not a metric unless they name how the query set is built and how often it is run.
- Ask what they would do first. Content volume as the opening move, before the technical floor and the footprint are dealt with, predicts a year of work that never compounds.
- Ask whether they will fix your directory and profile listings. It is unglamorous, hard to bill as a recurring retainer, and among the highest-leverage work available.
- Ask what happens if AI search does not send you traffic. An honest answer accepts that the effect surfaces first as branded search and direct visits, not as a referral line in analytics.