An AI SEO agency is hired to increase how often AI assistants name your brand when buyers ask them questions. The credible version of the work is three things: measuring which brands the engines currently name for your buying prompts, earning third-party mentions and coverage on the sources those engines lean on, and re-measuring on a fixed cadence to see whether the mention rate moved. Published 2026 retainers cluster between roughly $1,500 and $10,000 a month for small and mid-market scopes, with enterprise programmes quoted at $25,000 and up.
The term covers two very different products sold under one label. One is a genuine measurement-and-earned-media practice. The other is a conventional SEO retainer with a new cover page. This guide is written for the buyer, not the seller, so it deals mostly in how to tell them apart.
What the work actually consists of
Strip away the positioning and a real engagement has four moving parts.
Prompt research. Deciding which questions matter. Not keywords, questions: the phrasings a buyer actually types into an assistant when they are close to choosing. This set is the unit of measurement for everything that follows, so it should be built with you and written down.
Baseline measurement. Running that prompt set across the engines your buyers use and recording which brands get named, how often, and which sources the answers cite. Without this you cannot tell later whether anything worked. Our walkthrough of how to measure AI share of voice covers the mechanics.
Earned-media and source work. This is where the budget goes. Ahrefs' analysis found AI visibility correlates most strongly with third-party mentions and video, not with on-page work. That points the effort at review sites, comparison round-ups, community threads, podcasts and YouTube rather than at another 2,000-word blog post. Semrush found Reddit alone accounts for 40.1% of all AI citations, the single most-cited source.
Re-measurement. Rerunning the same prompts on a schedule and reporting the delta. If this is missing, you are buying activity, not outcomes.
If your agency's scope of work does not name all four, ask which one they left out and why. For a fuller breakdown of scope, see what an AI SEO service actually includes, or start further back with what AI SEO is.
What it costs
Very few agencies publish rates, which is itself a signal about the market's maturity. Below are the figures that were publicly listed on 20 July 2026. Every row is attributed. Where a band was not published, it is left out rather than filled in.
| Source | Scope | Published price |
|---|---|---|
| WebFX pricing guide | Small business, basic strategy | $1,500 to $5,000/mo |
| WebFX pricing guide | Medium business, advanced strategy | $5,000 to $25,000+/mo |
| WebFX pricing guide | Enterprise, complex strategy | $25,000 to $50,000+/mo |
| WebFX pricing guide | Project-based engagement | $5,000 to $50,000 per project |
| WebFX pricing guide | Hourly consulting | $50 to $300/hr |
| WebFX own GEO service | Listed entry price | From $3,000/mo |
| Percepture published packages | Launch / Growth / Authority | $3,500 / $6,000 / $8,500 per mo |
| Percepture published packages | Enterprise | $10,000+/mo |
| Percepture add-on | Digital PR activation | From $4,500/mo |
| The Digital Elevator guide | Monitor and maintain tier | $1,000 to $2,500/mo |
| The Digital Elevator guide | Active optimisation tier | $3,000 to $8,000/mo |
| The Digital Elevator guide | Category leadership tier | $10,000 to $25,000+/mo |
Two things stand out. The bands overlap heavily, meaning the price tells you almost nothing about the scope on its own. And the bottom tier, roughly $1,000 to $2,500 a month, is described by the agencies themselves as monitoring plus light reporting. That is software plus a slide deck. Priced against tools, where entry points run from a free check to about $29 a month for Otterly, $55 for SE Ranking, $89 for Peec, $295 for AthenaHQ and around $399 for Profound, a monitoring-only retainer carries a heavy margin. We break the software side down separately in AI visibility software pricing.
For context on the buyer side, "ai seo agency" gets about 1,000 US searches a month at a keyword difficulty of 8, but a cost per click of $82.43 (DataForSEO, July 2026). A high CPC against low difficulty means the demand is small, wealthy and heavily bid on. Expect to be sold to aggressively.
The deliverables to demand
Put these in the contract, not the kickoff call.
A written prompt set. Twenty to fifty buyer questions, agreed by you, frozen for the term. If the prompts change every month, the trend line is meaningless.
A baseline report with raw answers attached. Not a score. The actual text of what each engine said, with dates and the brands it named. You should be able to reread it in six months.
Named engine coverage. ChatGPT, Gemini, Perplexity and Claude behave differently enough that "AI search" as a single number hides more than it shows. Note also that ChatGPT and Gemini often hide their citations while Perplexity and Claude expose theirs, so citation-level reporting is only possible on some engines.
A fixed re-measurement cadence. Weekly or fortnightly, same prompts, same method, run at a comparable time. Monthly is acceptable. On demand is not.
Run counts per measurement. A single run per prompt is noise. Three or more runs per prompt per engine is the minimum for a defensible number.
Want to see this in action?
See how every major AI model talks about your brand. Free to start.
An earned-media log. Which placements, mentions, threads or videos were pursued, which landed, and on which dates. This is the part you are actually paying for and it is the easiest to leave vague.
Reporting that shows misses. A report where the line only goes up is a marketing document.
Why any promise of a stable rank is overselling
AI answers are not a ranking system, and this is the single most useful thing a buyer can understand before signing. SparkToro found the same AI query changes its answer roughly 70% of the time. Our own measurement backs that up.
On 13 July 2026 we ran 20 buyer prompts three times each across four engines via API, 240 answers in total. Engines named between 4.8 and 5.2 brands per answer. The top-ranked brand changed between identical runs at these rates:
| Engine | Top brand changed between identical runs | Brand-set overlap between runs |
|---|---|---|
| Gemini | 44% | 54% |
| Perplexity | 43% | 61% |
| ChatGPT | 35% | 42% |
| Claude | 28% | 67% |
Top-pick change rate
How often the top recommendation changes between identical runs
Cross-engine agreement is worse than within-engine noise. The same prompt produced the same top brand on only 20% of ChatGPT and Perplexity pairs, and 53% of Gemini and Claude pairs. Citation behaviour diverges just as sharply: ChatGPT cited 445 distinct domains, Claude 194 and Perplexity 142, and Forbes was the only domain appearing in all four engines' top citation lists.
The sources move too. Semrush recorded Reddit's share of ChatGPT citations falling from around 60% to roughly 10% inside a fortnight in late 2025.
So when an agency promises position one in ChatGPT, or a guaranteed rank in AI Overviews, the honest reading is that they are either not measuring repeatedly or not showing you the variance. The defensible metric is mention rate across many runs, tracked as a trend. Anything stated as a single position on a single day is a screenshot, not a result. Does generative engine optimisation actually work goes deeper on what does move.
The vetting checklist
Ask these on the first call. The right-hand column is the answer that should worry you.
| Question | Answer that should worry you |
|---|---|
| How many times do you run each prompt per measurement? | Once, or they do not know |
| Show me a client's baseline report with raw answers | They only have a dashboard screenshot |
| Which engines do you measure, and via API or manually? | "All of them", with no method named |
| What is your re-measurement cadence? | On request, or quarterly |
| What share of the retainer is earned media versus content? | Mostly on-page content production |
| Do you use a third-party tool, and can I see the bill? | Evasion, or the tool is the entire deliverable |
| What happens to my prompt set if I leave? | You do not get it |
| Can you guarantee a position in ChatGPT? | Yes |
| Which client of yours has this not worked for? | None, ever |
| How do you handle the 62% of AI citations that never name the brand? | Blank look |
That last one is worth expanding. Semrush found 62% of AI citations do not name the brand being cited, which means a large slice of the influence on an answer is invisible to link-based reporting. An agency that has thought seriously about AI visibility will have an opinion on this. One that has rebranded a link-building retainer usually will not.
When you do not need an agency
For a good number of teams, a tool plus a few hours a month of in-house effort replaces the retainer outright. The measurement half of the work is genuinely automatable, and the earned-media half is often better done by someone who already knows your customers and your product.
The rough test is scope. If you sell one product into one market, you can run a prompt set yourself and spend the difference on getting mentioned in the places the engines already cite. If you have multiple brands, several markets or a regulated category where every external mention needs review, the coordination cost is real and an agency starts to earn its fee. We laid the decision out in full in GEO agency or in-house, and the honest cost case sits in is answer engine optimisation worth it.
One caveat on incentives. We build Honeyb (our product), an AI visibility measurement tool, so we have an obvious interest in the tool-plus-in-house path. Weigh that. The measurement categories in this guide apply whether you buy them from us, from another vendor, or as part of a retainer.
Start with a baseline you own
Whatever you decide, get a baseline before you sign anything. It costs nothing to know which brands the engines currently name for your category, and walking into a pitch already holding that data changes the conversation. Run a free check at /tools/ai-visibility-checker and see where you stand today.
