Generative Engine Optimisation (GEO and AEO)
Making your organisation the source an AI answer draws on, through the mechanisms that plausibly drive citation — and telling you clearly which of them we can measure and which we cannot.
This field has more confident advice than evidence. Nobody outside the model providers can see the ranking mechanism, the surfaces change without notice, and a great deal of published GEO guidance is inference presented as method. We would rather be useful about that than sell certainty we do not have.
Generative engine optimisation, also called answer engine optimisation, is the practice of increasing the likelihood that AI answer surfaces — assistants, AI search modes and chat interfaces — draw on and cite your content when answering questions in your domain.
What is different about being an answer source
Classic search optimisation competes for a position in a list a person will scan. Answer surfaces compose a response from several sources and cite some of them, which changes what is worth doing:
| Concern | Classic search | Answer surfaces |
|---|---|---|
| The unit that wins | A ranking page | A quotable passage inside a page |
| What the user sees | Your title and snippet | A synthesis, sometimes attributed |
| Why you are chosen | Relevance and authority signals | Extractability, corroboration, entity clarity |
| Traffic | A click | Often none; the value is being the cited source |
| Measurement | Rank, impressions, clicks | Sampled prompts, mention share, referral traffic |
| Volatility | Ranking changes | Model updates change behaviour with no notice |
The traffic row is the one that reorders marketing plans. If a meaningful share of your audience gets its answer without visiting anyone, the goal shifts from clicks toward being the source the answer is built from — and toward measuring that directly rather than through a traffic proxy that no longer reflects it.
What we actually do
Make passages quotable on their own
A claim stated plainly in one or two sentences, immediately under a heading that matches the question, with the qualifying detail after rather than before. Content that requires three paragraphs of context to make sense is hard to extract, and extractability is the one mechanism everyone in this field agrees on.
Answer the question that is actually asked
Prompts are longer and more conversational than queries, and frequently comparative or conditional. We build the real question set from your sales calls, support tickets and search console data rather than from a keyword tool, because the phrasing differs materially.
Make the entity unambiguous
Who you are, what you do, where you operate, and consistently across your site, your structured data, and the third-party sources that describe you. Entity confusion is a plausible and fixable reason for being absent from answers about your own category.
Get corroborated elsewhere
Answer surfaces draw on many sources, so being described consistently on sites other than your own matters more than in classic SEO. This is unglamorous work — directories, industry sources, documentation, comparison sites, communities — and it is closer to public relations than to technical optimisation.
Keep the technical basics right
Crawlable, fast, structured data that matches the visible content, no critical content locked behind script rendering. None of this is novel; all of it is a precondition, and a surprising number of sites fail it.
Decide what AI crawlers may access
Which crawlers you allow is a business decision, not a technical default. Blocking protects content and removes you from the answers; allowing does the reverse. We put the trade-off in front of you rather than choosing quietly.
On llms.txt, plainly
Adoption of the llms.txt convention has grown quickly and the evidence that AI crawlers actually use it is weak — Google has publicly said it is not needed. We will add one because it costs almost nothing and may help later, and we will not claim it produces measurable results. Anyone selling it as a lever is ahead of the evidence.
Measurement, and its limits
- A tracked prompt set. Questions your buyers actually ask, run on a schedule across the main surfaces, recording whether you appear and how you are described. See AI visibility monitoring.
- Share of mention against named competitors, which is more informative than your own trend alone.
- Referral traffic from AI surfaces, segmented where the analytics allow it. It undercounts, and it is directional rather than complete.
- Answer accuracy about you, because being cited incorrectly is a distinct and more urgent problem than not being cited.
- What we cannot measure, stated explicitly: attribution from an uncited answer, influence on a buyer who never clicked, and the reason behind any individual change.
Results are sampled and noisy, and model updates move them independently of anything you did. We report movement with that caveat attached rather than claiming credit for every rise.
How the engagement runs
The prompt set and a baseline come first, because otherwise nothing that follows can be evaluated.
Question set and baseline
Real buyer questions gathered from sales, support and search data; current visibility and accuracy baselined across surfaces.
Audit
Extractability, entity clarity, structured data, technical basics and third-party corroboration assessed.
Implementation
Content restructured for extraction, entity and structured data corrected, corroboration work begun, crawler policy decided.
Re-measurement
Prompt set re-run; movement reported against baseline with volatility acknowledged.
Handover
Monitoring, the prompt set and the working method handed to your team.
What you receive
Content built to be quoted, entity clarity fixed, and a monitoring set you keep.
Tracked prompt set
The questions your buyers ask, with baseline visibility and accuracy across surfaces.
Extractability audit and rework
Content restructured so claims stand alone and match the question asked.
Entity and structured data
Consistent description of who you are across your site, markup and third-party sources.
Corroboration plan
Where you should be described consistently off your own domain, with progress tracked.
Crawler access decision
The trade-off documented and the policy implemented as you decided it.
Monitoring and method
Handed over, including an honest statement of what cannot be attributed.
Is this the right engagement?
Worth being direct. Generative Engine Optimisation (GEO and AEO) is the wrong spend in some situations, and those are listed rather than buried.
Good fit if
- Buyers research your category through AI assistants before contacting anyone.
- You are absent from AI answers where competitors appear.
- AI surfaces describe your products or positioning incorrectly.
- Organic traffic is falling while brand demand is stable or growing.
- You want measurement rather than assertions about AI visibility.
Choose something else if
- Your buyers do not use AI surfaces to research this category.
- The site fails the technical basics; fix those first and much of this follows.
- You want guaranteed placement in AI answers, which nobody can honestly offer.
- The requirement is monitoring only. See AI visibility monitoring.
Frequently asked questions
Marked up with FAQPage schema so these answers can surface directly in search results and inside AI assistant responses.
What is generative engine optimisation?
The practice of increasing the likelihood that AI answer surfaces draw on and cite your content. It overlaps heavily with good SEO — crawlability, structured data, authority — and adds emphasis on passages that can be extracted and quoted on their own, unambiguous entity information, and being described consistently on sources other than your own site.
Is GEO different from SEO?
Less than the marketing around it suggests. The technical and authority foundations are the same, and a site that fails at SEO will fail at GEO. What genuinely differs is that the winning unit is a quotable passage rather than a ranking page, that corroboration across third-party sources matters more, and that the outcome is often a citation rather than a click.
Can you guarantee we appear in AI answers?
No, and nobody can. The ranking mechanisms are not published, they change without notice, and the same prompt can produce different answers. What we can do is address the mechanisms that plausibly drive citation, measure your presence against a tracked prompt set, and report movement honestly including when it moved for reasons unrelated to our work.
Should we create an llms.txt file?
Probably, because it costs almost nothing — and not because it demonstrably works. Adoption has risen sharply while evidence of AI crawlers actually using it remains weak, and Google has said publicly that it is not needed. We will add one and we will not present it as a lever that produces results.
Should we block AI crawlers?
It is a business decision with a real trade-off, and we put it in front of you rather than choosing quietly. Blocking protects your content from being used and removes you from the answers your buyers are reading. Allowing does the reverse. Publishers and vendors land in different places on this for good reasons.
Often paired with this
Most clients combine two or three engagements from the AI for Marketing and Growth pillar. These are the ones that most often run immediately before or after.
AI Visibility and Brand Monitoring in LLMs
Tracked prompts across surfaces, share of mention against competitors, and correction of inaccurate claims.
Read more →AI SEO Services
AI applied where it helps — clustering, gap analysis, internal linking, log analysis — not to mass-produce pages.
Read more →AI Content Production Systems
Source-grounded drafting, brand voice from your own archive, and review gates that cannot be skipped.
Read more →Is this the right engagement?
Tell us what you are trying to build. If a different service fits better, or if you do not need us at all, we will say so.