What is actually being sold
An LLM SEO service sells you visibility inside the answers AI assistants give. The category is young enough that the same offer appears under several names: LLM SEO, GEO agency, AI visibility service, answer engine optimization consultancy. The label tells you almost nothing about what is included.
Underneath, real engagements tend to contain some combination of four things: measurement, technical work on your site, content, and presence in places you do not own. Those are very different in cost, in speed, and in how much anyone can guarantee.
The reason this page exists is that the newness of the vocabulary makes it unusually hard to tell a substantial offer from a familiar one with new nouns. Both are being sold, sometimes by the same people.
Ask which of the four things you are buying, and how many hours go to each. A supplier who cannot answer that is selling a category, not a service.
What the work genuinely consists of
Measurement
Building a question set that matches how your buyers actually speak, putting it to several assistants on a schedule, and recording who gets named. Fast to set up, genuinely useful, and changes nothing by itself. Any engagement should start here, because without a baseline nobody can tell later whether the rest worked.
Technical access
Confirming the AI crawlers can read your site, which comes down to what your robots.txt allows, whether a content delivery network is challenging them, and whether your pages render their content without JavaScript. Instant, binary, and blocks everything else if it fails. This is a couple of hours of work and it should not be presented as a programme.
Content
Making pages that answer the questions your buyers ask, with claims specific enough to quote. This is most of the billable time in most engagements and it is the part where quality varies most.
Presence you do not own
Being genuinely useful in the discussions and roundups that already get cited for your category. The slowest, the least automatable, and often the highest-value. It is also the part most likely to be quietly dropped from a proposal, because it does not scale.
What nobody can promise you
Three claims worth treating as disqualifying, because they describe things outside anyone's control.
A guaranteed position in an answer. There is no position. A model names three or four sources and composes a sentence. Nobody can guarantee inclusion, and the answer changes between runs of the identical question for reasons that have nothing to do with your site.
A specific timeline for a specific engine. Perplexity retrieves live and can reflect a new page within days. ChatGPT tends to move over ten to fourteen weeks. AI Overviews move when Google rankings move. Anyone quoting one number for all three has not looked.
Attribution to revenue. The path from a recommendation to a purchase almost always runs through a separate branded search or a direct visit, carrying nothing that identifies where it came from. A supplier promising to show you revenue from AI mentions is modelling, and should say so.
What can honestly be promised is the work itself, a measurement that shows whether it moved anything, and a realistic account of which levers are slow.
Telling a real offer from a rebrand
A large amount of what is sold as LLM SEO is competent search work with the nouns changed. That is not fraud and it is often useful, but you should know which you are buying.
Ask what they would do differently for this than for search. The honest answer is a short list: write to be quotable rather than comprehensive, measure by asking rather than by traffic, and spend more effort on being referenced elsewhere. If the answer is a long list, ask for specifics. If the answer is nothing, at least the pricing should reflect that.
Ask how they measure. A supplier who reports AI visibility using referral traffic has not understood the mechanism, because most recommendations produce no click. Our page on AI brand mentions covers why the trail disappears.
Ask what happens in month one. A real answer starts with a crawler check and a baseline. A vague answer that starts with strategy usually means the first invoice arrives before anything is measured.
Ask who writes the question set. If they generate it from your website, it will reflect your marketing language rather than your buyer's uncertainty, and the whole engagement will be measured against the wrong thing.
Agency, freelancer, or doing it yourself
The right shape depends on which of the four kinds of work you actually need, which is why the question of who to hire cannot be answered before the measurement.
Doing it yourself is genuinely viable for the fast half. The crawler check is an afternoon. The baseline is an afternoon. Rewriting your existing pages to lead with the claim is a week and mostly editorial judgement rather than technical skill. A small business can get a long way on this alone, and should, before paying anyone.
A freelancer suits the content half. If the measurement says you have no page answering the questions your buyers ask, that is a writing problem, and writing is straightforwardly buyable.
An agency earns its fee on the slow half. Building genuine depth, and getting referenced in the places that already get cited, is sustained work with relationships attached. That is the part hardest to do alone and hardest to buy well.
Nobody should be sold a retainer before a baseline exists. Without one there is no way to distinguish work that helped from work that happened, which suits the supplier and nobody else.
What this costs, and what drives the price
Pricing is unsettled and the headline figure rarely reflects what drives the bill.
Measurement scales with questions times engines times frequency. Twenty questions across four assistants weekly is a few hundred answers a month. Two hundred questions daily is tens of thousands. Those are different products at the same nominal retainer, and competitor tracking multiplies it again.
Content is the largest line in most engagements, and the one where you should ask for samples of work already published rather than a description of a process.
The slow work is priced by time and cannot be shortcut. Anyone offering to accelerate it is usually offering to buy something, and bought mentions in low-quality places do not get cited.
The cost that catches people out is the work the measurement creates. A service that tells you monthly which questions you are absent from is generating a content backlog. Budget for the writing or the retainer buys you a recurring reminder of a problem nobody is addressing.
Buy the measurement first, do the fast half yourself, and only then decide what is worth paying somebody to do. Any order other than that is buying before knowing.
How to brief a supplier so the work is judgeable
Most disappointing engagements in this category were badly briefed rather than badly executed, and the fix is unglamorous.
Give them the questions, or write them together. This is the single most useful thing a client can do, and it costs an afternoon. You know how your buyers speak and a supplier does not, and a set generated from your website will measure your marketing language rather than your market.
Name what you would consider success, in advance and in writing. Not increased visibility. Something checkable: named in at least a third of the shortlist questions in one named assistant by a named date. It may turn out to be the wrong target, and having a wrong target you can test beats a vague one you cannot.
Say which half you are buying. If you want the fast work, say so and expect it to be finished in weeks rather than retained. If you want the slow work, say so and accept that the first quarter will look like nothing is happening, because that is what building depth looks like from outside.
Ask for the baseline before any other deliverable. Everything else is unjudgeable without it, and a supplier who resists is telling you something about how they expect to be assessed.
A brief containing those four things makes a good supplier better and makes a weak one visible within a month, which is the point.
The things worth walking away from
Five, in rough order of how much they should worry you.
A guarantee of being named. Covered above. Nobody controls this and offering to is either misunderstanding or misrepresenting.
Reporting that leads with traffic. If the monthly report opens with sessions from AI referrers, the supplier is measuring clicking rather than visibility, and will report failure during a period of genuine progress.
A refusal to show the question set. The set determines everything the engagement is judged on. Not being allowed to see it is not a reasonable commercial position.
Volume promises. Any number of pages per month as the headline deliverable is a promise about output rather than outcome, and this is precisely the category where publishing width instead of depth fails hardest.
No mention of the slow half at all. A proposal that is entirely on-site work is a proposal that has quietly omitted the part which usually decides whether any of it gets retrieved. Our guide to what actually moves visibility separates the two.
What a monthly report should contain
The report is where an engagement is either honest or decorative, and it is worth agreeing its shape before signing rather than discovering it in month two.
The same question set, every month, unchanged. If the questions move, the history is worthless. Any change to the set should be flagged explicitly as the start of a new measurement rather than folded silently into a trend line.
Results split by assistant, not blended. A single number hides the case where you are gaining in one and absent from another, which is usually the most actionable thing in the whole report.
Every business named, not only yours. That set is your real competitive position and it frequently contains names nobody expected. When your own line moves for no reason you can identify, this is where the explanation usually is.
Some actual sentences. A handful of full answers, quoted. Being named dismissively counts as present in any tally, and reading the wording is the only way to catch a category quietly deciding you are the cheap option.
An honest note on what did not move. A report with no negative findings in a field this slow is a report that has been curated. The absence of bad news over several months is itself a warning sign.
One structural thing to agree at the start: what the report will look like in the months when nothing has happened. There will be several, particularly in the first quarter, and the pressure on a supplier to fill them with activity is considerable. A report that lists work completed rather than outcomes observed is easy to produce every month indefinitely, and tells you nothing about whether any of it is working. Agreeing in advance that a quiet month gets a short, honest report is a small conversation that prevents a slow drift into theatre.
Worth asking, too, what happens to the raw data if the relationship ends. The question-and-answer history is the genuinely valuable artefact, far more than the charts built on it, and it is the thing that cannot be reconstructed. If it lives only in a supplier's dashboard, a year of measurement disappears when you leave, and the next supplier starts from zero.
A note if you are on the other side of this
Worth saying, because the pressures in a young category push everyone the same way.
The commercial incentive is to promise the outcome, because the outcome is what the client wants to buy. The problem is that the outcome depends on a distribution nobody controls, on a timescale that is often longer than the contract. Promising it means either getting lucky or explaining later.
The alternative that holds up is to sell the work, the measurement, and an honest account of which levers are slow. It is harder to sell against somebody guaranteeing results, and it is the only version that survives twelve months of being checked.
The measurement is what makes that possible. A supplier who takes a real baseline in week one can show, a quarter later, exactly what moved and what did not, including the parts that did not work. That is a much stronger position than it sounds, because almost nobody in this category can currently do it.