LLM SEO agency, or the SEO agency you already have
An LLM SEO agency is hired to get a brand named inside AI answers, and most of what it should be doing is ordinary search work aimed at a different consumer, because a page an engine cannot crawl, index or rank is a page no answer can retrieve; what is genuinely new is the structural rewriting that makes a passage quotable and the measurement layer that watches a probabilistic output instead of a ranked list.
Key takeaways
- Retrieval leans on the same index search does. Unindexed pages cannot be cited, which makes indexation the floor of any LLM SEO engagement rather than an afterthought.
- One client we audited had 26 of 52 pages indexed and only 3 of 22 articles. AI Overviews fired on 4 of 4 buyer questions we tested, they had written articles for two of those questions, and neither article was indexed.
- The genuinely new work is structural: making the answering sentence exist as a clean passage, in the shape the engine tends to quote for that question.
- The other genuinely new work is instrumentation. Buyers type sentences into an LLM, and no keyword tool can see them. Five terms this category treats as obvious returned no volume data at all in our own August 2026 pass.
- Schema and llms.txt are hygiene with experimental upside. No controlled test of either has been published by anyone, including us, and selling one as a proven citation driver is not honest.
How much of LLM SEO is just SEO
More than the category likes to admit, and less than the sceptics claim. When an engine answers a commercial question, it does not consult a private library of brand knowledge. It runs searches, reads what comes back, and writes an answer over those sources. That makes ranking, indexation and crawlability inputs to citation rather than a parallel discipline. If your page is not in the index, no amount of AI-specific work puts it in an answer.
The market itself keeps admitting this by accident. Search demand for the term is falling even as the category grows: our own August 2026 measurement pass put llm seo at 880 searches a month, declining across the year from 1,300 to 480, while ai search optimization grew from 880 to 1,600 and aeo services grew from 140 to 720. The offer is stable, the vocabulary is churning, and one provider we inventoried publishes three separate service pages for a single AI-search offering, one per search term. That is a naming problem, not three services.
So a buyer evaluating an LLM SEO agency should expect the scope of work to be about two thirds recognisable. Crawler access, indexation, internal linking, entity clarity, third-party corroboration, comparison and cornerstone content. If a provider's scope contains none of that and is entirely AI-specific novelty, they are selling the roof of a house with no foundation.
Much of the underlying work overlaps with search optimization. Specialist providers add tracking of brand mentions across AI assistants. Provider A and Provider B both position around measurement, while Provider C emphasises content production.
illustrative, not a capture An answer at this level cannot tell you which half of the problem you have. If your pages are not indexed, the specialist is the wrong first purchase and the overlap argument is not a nuance, it is the whole decision. Written to show the shape of an engine answer. It has no session, no capture date and no sample size, so it is not evidence about any category, including yours.
The indexation floor, with a worked example
Here is what that looks like in practice on a real account. In August 2026 we audited an audio-visual production company in the San Francisco Bay Area, checking indexation page by page against Search Console rather than trusting a crawler's opinion. The result: 26 of 52 pages indexed, and only 3 of 22 articles.
Separately we tested buyer questions in their category and found AI Overviews firing on 4 of 4 of them, with no competitor holding the ground. The client had already written articles addressing two of those four questions. Neither article was indexed.
That is the whole argument in one account. They did not have an AI visibility problem that needed an AI-specific solution. They had 19 articles Google had never accepted into its index, which meant the content existed for humans who found it and for nobody else. Buying an LLM SEO retainer before fixing that would have been buying the second job while the first was still open, and any provider who sold it that way would have been selling the more expensive of the two problems.
The general rule is boring and it holds: check whether you are retrievable before you spend anything on being quotable. Retrievability is cheap to test and cheap to fix relative to a content programme, and it is the constraint far more often than the category's marketing suggests.
The part that genuinely is not SEO
Two things in this work are new, and they are the two a traditional SEO retainer will not cover by default.
The first is structural. Being in the pool of sources an engine reads is not the same as being the one it quotes. Selection is governed by whether the sentence that answers the question exists on your page as a clean, self-contained passage, whether your page is the shape the engine tends to quote for that question, and whether your claim agrees with the other sources it is reading beside you. Fixing that is rewriting, not link building, and it is the work most likely to be skipped by a retainer built around rankings.
The second is instrumentation, and it is the larger gap. A buyer asking an engine which company to hire types a sentence, not a keyword, and no volume tool can see those sentences. Our own August 2026 pass through a keyword data provider is the demonstration: five terms this category treats as obvious returned no data at all. Not low volume, no data. Those terms were aeo agency, llm seo agency, answer engine optimization agency, generative engine optimization agency and ai visibility agency. This page targets one of them deliberately, and by the standard instrument it addresses a market that does not exist.
The instrumentation gap runs deeper than keyword tools. On another account we found that the standard Search Console summary export caps at 1,330 queries while the full pull returned 5,983 for the same property and window. Roughly four in five appearances were invisible in the export most agencies read from. If a provider is diagnosing your AI visibility from a dashboard that truncates by default, the diagnosis is built on the fifth of your surface the tool happened to show them.
Schema, llms.txt, and the levers nobody has tested
Most LLM SEO pitches converge on a short list of AI-specific artefacts: schema markup, an llms.txt file, agent-readable pricing and fact pages. We ship all of them. We also test them, and we will not describe any of them to a client as a proven citation driver, because no controlled test of any of them has been published by anyone, us included.
The honest framing is that they are entity hygiene with experimental upside. Structured data helps machines agree on who you are and what you sell, which is useful whether or not it moves a citation. A curated machine-readable index of your own pages costs an afternoon and cannot hurt. Neither has a demonstrated causal link to being named in an answer, and the absence of that evidence is an opportunity for whoever runs the first clean test, not a licence to bill the hypothesis as a mechanism.
There is one place where schema is not optional, and it is a rule rather than a theory. Every answer you put into markup has to exist as readable text on the page. Markup-only content is discounted and it breaches the published guideline. If you cannot point at the sentence on the page, it does not go in the markup, and the fix is to add the visible text rather than to delete the schema. We check this on every page we ship, because we have found live client pages carrying FAQ answers that appeared nowhere in the visible copy at all.
Whether your current SEO agency should just do this
Often, yes, for the first half of it. If your agency is already fixing indexation, earning corroboration and writing pages that answer questions directly, they are doing most of the retrievability work whether or not they call it LLM SEO. Adding a second vendor to duplicate that is paying twice for one job.
What they will usually not have is the measurement layer, and the reason is structural rather than a failure of skill. We fetched the homepages of two large marketing incumbents live on 2026-08-15. One lists roughly eighteen services and mentions AI search, LLM visibility, answer engines and citations nowhere; its only AI is a chat widget, which is a conversion tool. The other splits its offering into technology, advertising and marketing with no AI-search line at all. These are not small or bad firms. They simply do not sell this, so they do not measure it, and a channel nobody measures is a channel nobody is accountable for.
That points at the sensible shape of a first engagement: a measurement layer over the work you already buy, rather than a replacement for the vendor you already have. It removes the disruption objection entirely, because nothing about measuring AI visibility conflicts with an existing search retainer, and it means the question you are answering is who is watching this channel rather than whether to fire anyone.
Where we are the wrong call. If your pages are not indexed, your first purchase is a technical fix and it is cheaper than us; we will tell you that on the free assessment and it costs you nothing to hear. If you want a self-serve dashboard your team runs, one of the software providers is a better fit and one of them publishes a from-$6,000 monthly price for a platform with a strategist attached. If you need a case study from your exact vertical with revenue attached to it, we do not have one: we have no revenue-attributed win, we hold every client citation share figure we have measured because they are single-sample readings with roughly 25 points of margin, and we would rather say that than show you a number that will not survive being re-run.
Questions, answered plainly.
Is LLM SEO just SEO with a new name?
About two thirds of it is. Engines answer commercial questions by running searches and writing over what comes back, so ranking, indexation and crawlability are inputs to citation rather than a separate discipline. The genuinely new parts are structural rewriting, so the answering sentence exists as a clean passage, and instrumentation, because the questions buyers type into an assistant are invisible to keyword tools.
Can an unindexed page be cited by an AI assistant?
Not through the retrieval paths we measure. That makes indexation the floor of the engagement. On one account we audited in August 2026, 26 of 52 pages were indexed and only 3 of 22 articles. AI Overviews fired on 4 of 4 buyer questions we tested, the client had written articles for two of those questions, and neither was indexed.
Does llms.txt or schema markup get you cited?
Nobody has published a controlled test, including us. We ship both and we test both, and we describe them as entity hygiene with experimental upside rather than as citation drivers. One rule is not optional: every answer you put into markup must exist as readable text on the page, because markup-only content is discounted and it breaches the published guideline.
Should I hire a separate LLM SEO agency or extend my current retainer?
If your current agency already fixes indexation, earns corroboration and writes pages that answer questions directly, they are doing most of the retrievability work already and a second vendor duplicates it. What they usually lack is the measurement layer. The sensible first engagement is a measurement layer over the work you already buy, which conflicts with nothing and makes the channel accountable to someone.
Why does llm seo show falling search volume if AI search is growing?
Because the vocabulary is churning while the offer stays the same. Our own August 2026 pass put llm seo at 880 a month, declining across the year from 1,300 to 480, while ai search optimization grew from 880 to 1,600 and aeo services grew from 140 to 720. Volume tools measure what people type into Google, which is a poor proxy for what they ask an assistant.
What is measured, and what this page is not.
This is an explainer. It carries no figures, and it is not a reading of your category. The disclosure below states the instrument that produces the numbers the essay refers to, so the distinction is on the page rather than assumed.
- instrument
- Caul
- what was measured
- Nothing on this page. Where the essay refers to citation share, that figure is produced separately, per account.
- how
- A prompt set written once for a category and then frozen, run against every engine in clean sessions, with each answer stored unmodified.
- over what window
- Reviewed on 2026-08-16. The engines change, so read the essay against the date on the byline.
- what this cannot tell you
- An explainer is not evidence about your category. Being named is not being recommended, and it is not traffic or revenue. Any figure about your own visibility has to come from a capture of your own category, carrying its sample size and its window.
The rest of the cluster.
See where you stand.
The audit is a real sweep of your category, benchmarked against competitors you name, delivered on a call so the findings get explained rather than emailed. You keep the report and the underlying data whatever you decide afterwards.