AI SEO
Getting cited by the models, not just ranked by Google.
People increasingly ask ChatGPT, Perplexity and Google's AI Overviews the questions they used to type into a search box, and the answer names a handful of sources. Most of what makes you one of them is unglamorous and already familiar: be crawlable, be unambiguous about who you are, and write something concrete enough to be worth quoting.
Straight answer
Most of AI SEO is technical SEO with a new name. A little of it is genuinely new.
We would rather tell you that than sell you a new discipline. The work that moves AI visibility is mostly work that was already worth doing, plus a small set of things that are actually new and cheap to add.
What is being sold
- Guaranteed placement in AI answers
- llms.txt described as a ranking factor
- "AI optimised" content that reads like every other page
- A brand new discipline with a brand new retainer
- Dashboards tracking metrics nobody can influence
What actually influences it
- An unambiguous entity: who you are, stated consistently
- Clean structured data a model can resolve without guessing
- Passages that stand alone and are therefore quotable
- Being mentioned and cited on sites beyond your own
- A deliberate crawler policy rather than an accidental one
Said plainly
AI answers are non-deterministic and no provider publishes a ranking mechanism, so nobody can promise you a citation. What can be done is concrete: make the site machine-readable, make the entity data unambiguous, make the writing quotable, then measure whether you actually appear for the prompts that matter. If your site is not yet crawlable and clearly structured, that is the work, and it is the same work it always was.
Scope
What an AI SEO agency can genuinely do.
Six things, in rough order of how much they matter. The first three are foundations that also help ordinary search, which is exactly why they are worth doing.
Entity clarity
Who you are, what you do and where, declared consistently in schema and repeated the same way everywhere you appear. Models confuse similarly named organisations constantly, and an unambiguous entity graph is the single best defence against being described as someone else.
Structured data that resolves
A connected graph using @id references rather than a pile of disconnected blocks, so a machine can follow the relationships between organisation, person, service and article without inferring them. This is ordinary technical work done properly.
Quotable content structure
Self-contained passages that answer one question completely, so an extract still makes sense with the page removed. Concrete claims with numbers get quoted; hedged marketing prose does not, because there is nothing in it worth lifting.
Machine-readable indexes
llms.txt and llms-full.txt, generated from the site rather than hand-maintained so they cannot drift, alongside clean XML sitemaps with honest last-modified dates. Cheap to publish, and sold to you as exactly what it is rather than as a ranking factor.
Off-site presence
What the rest of the web says about you, because that is a large part of what a model has learned. Consistent entity details across the sites that mention you, and earned coverage that puts your name near the topic you want to be associated with.
Crawler policy
A deliberate decision about which AI crawlers may read your site, separating the retrieval bots that cite you from the bulk training bots that do not. Set to match your commercial intent instead of inherited from a plugin default.
How it runs
Baseline, fix, publish, measure.
Prompt baseline
We agree the questions a real buyer would ask a model about your category, run them, and record who gets named today. Without that baseline any later claim of improvement is unfalsifiable, which is how most of this is currently sold.
Usually two to three daysFix the foundations
Entity and schema work, crawlability, and the content restructuring that makes passages quotable. Then the machine-readable indexes, generated from the site so they stay accurate on their own.
Scoped from the baselineRe-measure honestly
The same prompts re-run on a schedule, reported with the movement and the noise, including where nothing changed. AI answers vary run to run, and a report that hides that is not a report.
Logged as it is spentEvidence
We did this to our own site first.
The fastest way to judge whether someone can do this work is to look at whether they have done it where it costs them something.
Published
A machine-readable index, generated not hand-written
This site publishes llms.txt and a fuller llms-full.txt, plus a browsable version for people. All three are generated from one catalogue at build time and automatically pick up new pages, so they cannot quietly drift out of date.
Verified
A connected entity graph, not a pile of blocks
Every page carries structured data that validates, using @id references so person, service, article and breadcrumb resolve as one graph. We have also fixed the opposite on client sites: duplicate organisation entities and invalid markup competing with each other.
Cost
What it costs.
AI SEO is billed at £90 to £110 an hour, or the £95 blended rate where it crosses into development. There is no separate AI premium, because most of the work is technical SEO. Minimum engagement is 4 hours.
Visibility baseline
From £450
An agreed prompt set run across the main assistants, recording who is named today and which sources they pull from, so any later change is measurable rather than asserted.
Foundations build
Scoped
Entity and schema graph, crawlability, quotable content structure, generated llms.txt and llms-full.txt, and a crawler policy that matches your commercial intent.
Tracked retainer
From £600/mo
The prompt set re-run and reported on a schedule, with the ongoing entity and content work. Cancel any time, and we will tell you when it has stopped being worth paying for.
Time is logged as it is spent and the ledger is open to you, including anything scrapped or reverted. Full breakdown on the rates page.
FAQ
Questions clients actually ask
Is AI SEO actually a different discipline, or a rebrand?
Mostly a rebrand, and anyone telling you otherwise is selling something. The models are trained and grounded on the same web Google crawls, so the fundamentals do not change: be crawlable, be unambiguous about who you are, and say something worth quoting. A site that is genuinely good at technical SEO is already most of the way there.
What is genuinely new is narrow, and worth doing: making your content extractable in self-contained chunks, being explicit about entities so a model does not confuse you with someone else, and publishing a machine-readable index rather than making a model parse your rendered HTML. That is the honest scope of this page.
What is llms.txt and does it actually do anything yet?
llms.txt is a proposed convention: a plain-text, markdown-structured index at the root of your site that tells a language model what the site covers and where the important pages are, without it having to crawl and strip HTML. There is also an llms-full.txt convention carrying a fuller summary per page.
The honest position: no major model provider has committed to reading it, so treating it as a ranking factor would be dishonest. It costs very little to publish, it makes your content easier to ingest correctly if it is read, and it forces you to write a clear one-line description of every page, which is useful regardless. We publish ours, and you can see the browsable version. We would not sell it as more than that.
How do I get cited by ChatGPT, Perplexity or AI Overviews?
Broadly by being the clearest available source for a specific claim. Models grounding an answer pull from pages that state something concrete, attributably, in a passage that stands on its own without the surrounding page. Vague, hedged marketing copy is almost never quoted because there is nothing in it to quote.
The other half is being mentioned elsewhere. Models weigh what the wider web says about you, so earned coverage and consistent entity information across the sites that mention you matter as much as anything on your own domain.
Can you guarantee my brand appears in AI answers?
No, and neither can anyone else. AI answers are non-deterministic: the same question can return different sources on different days, and none of the providers expose a ranking mechanism you can optimise against the way you can with a search engine. Any guarantee of AI visibility is either a misunderstanding or a lie.
What we can do is measurable and honest: make the site machine-readable, make the entity data unambiguous, make the content quotable, and then track whether your brand actually shows up for the prompts that matter to you.
Should I block AI crawlers?
It depends what your content is for, and it is a genuine business decision rather than an SEO one. If your content is the product, for example a paid research library, blocking training crawlers is defensible. If your content exists to bring you customers, blocking the crawlers that ground AI answers removes you from an increasingly common way people find suppliers.
These are separable: you can allow the retrieval crawlers that cite you while disallowing bulk training crawlers. We will set the policy to match your intent rather than apply a default.
Keep reading
Related services
Book me
Do the models know who you are?
Send us your domain and the questions a buyer would ask an assistant about your category. We will run them, tell you who gets named instead of you, and whether that is fixable.