The three tiers, honestly
“AI visibility consultant,” “AEO agency,” “GEO consultant” — the labels blur, but you’re choosing between three genuinely different products:
| Tier | Typical 2026 price | What it is | Right for you when |
|---|---|---|---|
| Tracking tools (Otterly, Profound, Peec and similar) | $19–199/mo | Software that checks whether AI mentions you and shows trend lines | You want the number and you’ll do the work yourself |
| Consultant / done-for-you | ~$500 audit, $1,500–5,000/mo | A human who measures, diagnoses, executes the fixes, and re-measures | You want the outcome, not a second dashboard to check |
| Enterprise agencies | $5,000–30,000/mo | Teams, integrations, brand programs | You’re a multi-location brand with a marketing department |
The failure mode isn’t picking the wrong tier — it’s paying consultant prices for tool-tier work. A dashboard with a monthly call wrapped around it is still a dashboard.
What the real work actually is
The engines don’t take payment for placement, so a real consultant works the inputs the answers are built from. In our own cross-engine measurement, the things that repeatedly earn a small business its way into an AI answer are unglamorous: a page that answers the exact question in depth, concrete published facts the engine can quote (your price, your offer, your credential, stated in plain sentences with your name attached), a review layer on the sites the engines actually cite, and directory or registry corroboration for every claim. The engines increasingly tell users to verify credentials themselves — the work is making sure that verification finds you and confirms you.
Measurement has to bracket all of it. AI answers change run to run — in our durability study, an unmanaged first-place recommendation had an expected lifetime of roughly one to two days. A single screenshot proves nothing on either end of an engagement. Repeated, logged runs on frozen questions are the only honest before-and-after.
Five questions that expose a bad one
- “Show me a run log.” Not a score — the actual answers, timestamped, from a logged-out vantage. If the proof is a single number that only goes up, walk.
- “How many times do you run each question?” One run is a coin flip. Three-plus runs per engine per question is the floor for a claim.
- “Which engines, separately?” The engines disagree with each other more than they agree. A blended score across engines hides exactly the information you’re paying for.
- “What do you publish without my approval?” The right answer is nothing. AI-generated volume posted to your site under your name is a liability you’ll be cleaning up for years.
- “What happens when a position decays?” Positions rot. If there’s no re-measurement and re-hardening loop, you’re buying a one-time project with a subscription price.
The short version
Start with measurement — free or tool-tier. If the measurement shows you’re already the answer, defend it yourself and spend nothing. If it shows a gap on questions with money behind them, hire the tier that fixes causes and proves movement: published facts, answer pages, the cited review layer, re-measured on a log you can check.