AI Search Optimization Services
These services benchmark prompts, do technical setup (llms.txt, schema), optimise passages, earn citations from third-party sites, and report monthly. Only the fourth one moves the number.
The baseline nobody quotes: across 4,163 AI answers we ran, the tracked brand appeared in 4.8% of them. One answer in twenty is normal. Know that before you judge what a service is promising you.
The AI search optimisation market is about eighteen months old and growing fast enough that Google's own data shows searches for these services up several thousand percent year on year. That combination, new market plus urgent demand, is exactly where buyers get the least useful information.
We do not sell AI search optimisation as a service. We sell an outreach agent, and we run AI visibility scans inside it, which means we hold measurement data most buyers never see. That is what this page is for.
The baseline: what normal actually looks like
Almost every claim in this market is a relative one. "Double your AI visibility." "Get mentioned 3x more often." None of that means anything without a starting point, and nobody publishes a starting point.
Ours, across 4,163 answers covering 1,088 distinct tracked queries for 24 businesses between 8 July and 24 August 2026:
| Engine | Answers | Brand mentioned | Rate |
|---|---|---|---|
| ChatGPT | 1,157 | 58 | 5.0% |
| Gemini | 1,868 | 90 | 4.8% |
| Claude | 1,138 | 53 | 4.7% |
| All engines | 4,163 | 201 | 4.8% |
Two things follow from this, and both are useful when a salesperson is talking.
A brand appears in roughly one AI answer in twenty. So "we doubled their AI visibility" means two extra mentions in fifty answers. On a small prompt set that is inside normal variance. Ask how many answers the claim is based on before treating it as a result.
The engines are nearly identical. 5.0%, 4.8% and 4.7% is not a meaningful spread. A good deal of this market is sold on per-engine strategy, on the premise that ChatGPT and Gemini reward materially different things. On brand mentions, for the businesses we track, they do not. If a provider is selling separate optimisation tracks per engine, ask what difference they are claiming and how they measured it.
What these services actually do
The published service breakdowns are remarkably consistent. Page One Power, for instance, describes five steps: benchmarking target prompts, technical optimisation covering llms.txt and schema, passage-level content optimisation, earning links and citations, and monthly reporting.
| Step | What it is | Does it move the number? |
|---|---|---|
| Benchmarking prompts | Deciding what to track and taking a baseline. | No. It is measurement, and necessary. |
| Technical setup | llms.txt, schema, crawler access. | Only if something was broken. Mostly a one-off afternoon. |
| Content optimisation | Rewriting passages to be quotable. | Marginally. Helps once you are already being retrieved. |
| Citations and links | Earning coverage on third-party sites. | Yes. This is the lever. |
| Reporting | Monthly mention and citation tracking. | No. It tells you whether the rest worked. |
Notice what the load-bearing step is. Models mention brands they encountered in the sources they were trained on or retrieved from, which means the work is getting written about on sites those models draw from. That is outreach. It is the same job as getting cited by ChatGPT, and it is not a technical exercise.
This is worth knowing because the technical half is the visible half. A provider can show you a tidy llms.txt file and a schema audit in week one. Neither will change your mention rate.
Why nobody publishes a price
We checked. Page One Power gates its AI search optimisation pricing behind a form rather than publishing a rate. That pattern holds across most of the market as of August 2026.
The nearest public anchor is the same firms' link building pricing, which is the same underlying work. Published agency floors start around $2,500 a month, with Page One Power at roughly $3,500 to $3,700 and Siege Media carrying $5,000 minimums. Our full breakdown is in how much backlinks cost.
Use those as your sanity check. An AEO quote well below the link-building floor is either not doing the citation work, or doing it at a volume that will not move a 4.8% baseline.
The measurement trap
There is a specific way these engagements go wrong, and it is about how visibility is counted.
In the same dataset, the engines diverge sharply on whether they show sources at all:
| Engine | Answers | Cited a source URL | Rate |
|---|---|---|---|
| Gemini | 1,868 | 1,847 | 98.9% |
| Claude | 1,138 | 1,122 | 98.6% |
| ChatGPT | 1,157 | 852 | 73.6% |
Roughly a quarter of ChatGPT answers carry no attributable source. So a provider measuring your visibility by counting how often your domain appears in citation lists will systematically understate ChatGPT, because a quarter of the time there is no list to appear in. Ask whether they read the answer text for brand mentions or only parse citations. Those are different measurements and they will disagree.
What to ask before you sign
What is my baseline, measured over how many answers? If they cannot tell you where you start, they cannot tell you that you improved.
Is the prompt set fixed? A prompt list that changes between reports makes the trend meaningless. Comparability over time is the whole value of tracking.
Which model tier do you query? Almost nobody asks this. Running thousands of scheduled prompts on flagship models is expensive, so tools often use the cheap tiers. Our own figures come from the fast, low-cost tiers and we say so; a provider quietly doing the same is not measuring what your customers see.
How many placements a month, and on which sites? This is the step that moves the number. If the answer is vague, the engagement is mostly technical work and reporting.
Do you read the answer text or only the citations? Covered above. On ChatGPT it changes the result materially.
Agency, tool, or yourself
| Your situation | The honest answer |
|---|---|
| You have budget and no time | An agency. You are buying back hours, which is a real thing to buy. |
| You want to know where you stand first | Measure before you buy. Check it by hand or use a tracking tool. |
| Your constraint is budget, not knowledge | Run the outreach with software. The citation work is the same work. |
| Your llms.txt and schema are missing | Fix those yourself first with our free llms.txt generator and checker. It is an afternoon, not a retainer. |
| You are in a regulated or highly technical niche | A specialist who understands the field. Relevance judgement is the hard part. |
For the tooling side of this, best GEO tools compares fourteen products against real citation data, and LLM SEO tools explains what the tracking category actually measures. If the flat-fee route fits, our outreach agent runs the citation-earning half for $99 a month.
Frequently asked questions
What do AI search optimization services actually do?
Benchmark prompts, technical setup, content optimisation, earn citations from third-party sites, and report. The citation step is the one that moves the number; the rest is preparation and measurement.
How much do they cost?
Almost nobody publishes a rate. The nearest anchor is the same firms' link building pricing: from about $2,500 a month, $3,500 to $3,700 at Page One Power, $5,000 minimums at Siege Media.
What is a normal AI brand mention rate?
4.8% across our 4,163 answers, and near-identical by engine (ChatGPT 5.0%, Gemini 4.8%, Claude 4.7%). One answer in twenty is the normal starting point.
Do you need an agency for this?
Only if you want the outreach done for you. The technical pieces are a one-off afternoon; the citation work is ongoing and is what you are really paying for.
How should you measure whether it is working?
Fixed prompt set, mention rate read from the answer text rather than only citations, and a known model tier. ChatGPT cites a source only 73.6% of the time, so citation-only counting understates it.
Method
Visibility figures cover 4,163 successfully completed AI answers collected between 8 July and 24 August 2026, across 1,088 distinct tracked queries for 24 businesses, on ChatGPT, Gemini and Claude. A further 25 runs errored and are excluded. "Brand mentioned" means the tracked brand appeared in the answer text; "cited a source" means the response carried at least one source link. Each engine was queried on its fast, low-cost model tier rather than its flagship, which keeps scheduled scanning affordable but is a real limitation: flagship citation behaviour may differ and these numbers should not be read as describing it. Agency pricing was read from published vendor pages; where a provider gates pricing behind a form we say so rather than estimating.