Being ranked is not the same as being cited.
AI search optimisation, also called AEO or GEO, structures your content and entity so that AI answer engines cite you. Haryes builds the connected schema graph, writes answer-first content that retrieval systems can lift, and tracks your citation share across a set of tracked prompts.
Answer engines summarise. Someone gets named, and it may not be you.
When a potential customer asks ChatGPT, Perplexity or Google’s AI Overviews for a web developer in Nairobi, an answer is assembled from sources those systems can parse and trust. Ranking tenth for the query has no bearing on whether you appear in it.
Three things usually keep a business out of those answers.
- No clearly defined business entity: the schema is fragments on separate pages that never connect.
- Answers buried in the middle of long pages, in a form nothing can lift cleanly.
- Details that contradict each other across the site, the business profile and directories.
How it runs
Measure first, because without a baseline there is no way to tell whether any of this worked.
Build the prompt set
Fifty to a hundred prompts a real customer might ask, written from your enquiries rather than from a keyword tool.
A fixed measurement set.
Take a baseline
Run the set across the major answer engines and record where you are named, where you are linked and who is recommended instead.
A number to improve on.
Fix the entity
A connected schema graph resolving to one organisation, matched by consistent details across every profile and directory.
A business machines can identify.
Restructure the content
Answer-first openings, one question per page, and the specific local detail a summary cannot get anywhere else.
Pages worth lifting from.
Re-measure monthly
The same prompts, the same way, every month, reported as a trend rather than a snapshot.
Evidence it is moving.
What answer engines actually reward
Four things, none of them tricks. They are what makes a source safe to quote.
A resolvable entity
One organisation, described consistently, that the system can identify with confidence rather than guess at.
Confidence is the gate.
Liftable answers
A direct answer in the first fifty words that reads correctly on its own, out of context.
Write for the excerpt.
Specificity
Prices in shillings, named systems, local constraints. Vague copy cannot be cited because there is nothing to repeat.
Detail is citable.
Corroboration
The same facts on your profiles and listings, so nothing contradicts your own site.
Contradictions disqualify you.
What you get
Runs inside a monthly SEO plan, alongside technical and content work.
A citation baseline
Your starting position, measured.
- A fixed prompt set built from your enquiries
- Scored across the major answer engines
- Competitors recorded alongside you
Entity optimisation
One business, unambiguously.
- Connected schema graph
- Consistent details across profiles
- Contradictions found and corrected
Content restructuring
Written to be lifted.
- Answer-first openings
- One question per page
- Local specifics a summary cannot invent
Monthly tracking
The trend, not the snapshot.
- Citation share and link share
- Accuracy errors logged
- What changed and what we did
Questions clients ask before they commit
Can you guarantee we appear in AI answers?
No, and nobody can. These systems do not sell placement and do not publish their selection rules. What we can do is make you accurate, specific and structured enough to be a safe source, then measure whether citation share moves.
Is this different from normal SEO?
It overlaps heavily and the differences matter. Both want a technically sound site with good content. AI search adds entity clarity and a writing structure built for retrieval rather than for a results page.
How do we know it is working?
The tracked prompt set, run monthly. If citation share is flat after several months we say so and change the approach. The method is set out in our article on tracking AI citations.
Should we block AI crawlers instead?
That removes you from citations without bringing the clicks back. It is a reasonable choice for a publisher selling subscriptions and a poor one for a business that wants to be recommended.
Where to start
You do not know whether answer engines mention you at all.
We build the prompt set and take a baseline before changing anything.
Get a baselineThe site is healthy and you want the next thing.
Entity and retrieval work sits on top of what you already have.
Read the AI search insightsNobody can promise you a place in an answer. You can make yourself the obvious thing to cite, and then measure it.
