Selling AI visibility as an agency comes down to three decisions: package it as an audit-fix-monitor loop instead of a one-off report, price each tier against work your clients already buy, and prove every claim with evidence they can check themselves — the exact prompt, the exact engine, the exact answer. The fastest way to build that proof is a free AI visibility audit you run live with the client.
That formula sounds obvious. The strange part is that almost nobody is publishing it. In a July 2026 live probe, we ran "why is my competitor cited by ChatGPT and not me?" across three engines. Not one answer named a tool or vendor — the citations went to small personal blogs. The playbook for this whole space, including how to sell it, is unclaimed. The playbook for selling this service is unclaimed. This page is our attempt to write it down.
Why are clients suddenly asking about AI visibility?
Because their buyers stopped clicking. Because their buyers stopped clicking: a growing share of searches now ends inside a generated answer or on the engine's own properties, never reaching the open web. The rest end on Google's own properties or inside a generated answer. Clients feel this in their analytics before they can name it. Then someone on their team asks ChatGPT for a recommendation in their own category, watches a competitor get named, and calls their agency.
Here is the part clients cannot see, and the reason this service exists at all. Across the 2,994 probes we ran in July 2026, nearly one answer in three (30.1%) named brands without citing a single source: pure model memory. That is 900 answers where no page was fetched, no referrer was logged, and no analytics tool anywhere recorded an impression. A client can be losing recommendations every day with nothing in their dashboards to show for it.
This is why "we already do good SEO" no longer closes the conversation. The two disciplines optimize different machines; we walked through the mechanics in GEO vs SEO. The short version: an SEO loss costs positions, while a GEO loss means a competitor gets named every time the question is asked and your client never sees the impression they missed.
What does an AI visibility service actually include?
The service is a loop: audit where the client appears today, fix the gaps that keep them out of answers, and monitor monthly because answers move. Every deliverable you sell is a version of one of those three steps.
The audit answers three client-legible questions. Which buyer prompts matter in their category. Which brands ChatGPT actually names when those prompts run. Which sources it cites when it answers. Package the output as an appearance rate, a competitor leaderboard, and a citation-source breakdown — three artifacts a client can read without a glossary.
The fix work splits into on-page and off-page. On-page moves fast: in our crawl of pages that win AI citations, pages with FAQ blocks were more common among cited pages, pages served with and question-form headings roughly 2x. Those are afternoon-sized changes with measurable odds attached. Off-page is slower but heavier: Muck Rack's citation research found 84% of AI citations come from earned media, and Ahrefs' 75,000-brand numbers make the case for this lane: branded mentions correlate 0.664 with AI visibility; backlinks just 0.218.218 for backlinks, with YouTube mentions the strongest single signal at roughly 0.737. If your agency already does digital PR, you hold the highest-value input in this channel and most of your competitors do not.
One more finding changes who you can sell to. Industry analyses put roughly 90% of the pages ChatGPT cites at Google position 21 or beyond — domain authority is not the gate. That makes the offer credible for local and mid-market clients who could never win a head-to-head SEO fight against national incumbents.
The monitor step is what turns a project into a retainer, and it is defensible on the data. Semrush's September 2025 longitudinal test published 81 new pages and tracked them for 30 days: 42% were cited by ChatGPT within 30 days, Google AI Mode cited 36% on day one, and the rest mostly never got cited. Results arrive fast when they arrive at all, and they keep shifting afterward. A quarterly check misses both facts.
How should you package and price AI visibility services?
Three tiers: a one-off audit that opens the door, a monthly monitor-and-fix engagement that captures the recurring value, and a full GEO retainer for clients in genuinely contested categories.
The audit tier is the door-opener — cheap for the client to say yes to, cheap for you to produce, and it generates the fix list that justifies tier two. The monitoring tier is where the recurring revenue lives. The full retainer adds the off-page work: earned media, review-site presence, roundup placement. Listicles take roughly 41% of commercial-intent AI citations by industry citation analyses, so getting clients into the right roundups is retainer-grade work, not an afterthought.
| Tier | What's included | What the client sees |
|---|---|---|
| One-off audit | Category prompt set run on ChatGPT across three search modes; gap analysis; prioritized fix list | Appearance rate, competitor leaderboard, citation-source breakdown, one clear action plan |
| Monthly monitor + fix | Audit re-run monthly; on-page fixes (FAQ blocks, schema, question-form headings); change log tied to each re-run | Month-over-month appearance trend, before/after evidence for every fix shipped |
| Full GEO retainer | Everything above plus earned-media outreach, review-site and roundup placement, content built to be cited | Category share of voice across search modes, source-level wins, quarterly strategy review |
On pricing we will stay honest: there is no reliable published benchmark yet, so treat pricing as a positioning decision. What agencies typically do is anchor each tier to work the client already understands. The one-off audit prices like a technical SEO audit. The monitoring tier prices like a rank-tracking-plus-reporting engagement. The full retainer sits alongside content and digital-PR retainers, because the work overlaps heavily with both. And since the audit tooling itself can be free, the entry tier is almost pure analysis time — a low-risk opener for you and for the client.
How do you solve the proof problem?
Show receipts, not scores: the exact prompt, the engine, the date, the full answer, and the sources it cited. This is the single thing that separates agencies who keep these retainers from agencies who lose them at the first quarterly review.
Clients are getting skeptical, and they should be. Run three AI visibility tools on the same brand and you can get three different verdicts. That is not necessarily any tool lying; it is the medium. In our 300-probe phrasing study, two engine presets given identical questions agreed on category winners barely one time in five (22%). The engines are probabilistic. Sampling one once and reporting a single number is like polling one voter and publishing an election forecast.
If your deliverable is an opaque score, that variance destroys trust: the client checks a prompt themselves, sees a different answer, and now the whole report is suspect. If your deliverable is evidence, the same variance becomes your best sales argument — it is precisely why one-off checks are worthless and monthly monitoring is the product. Apply the same filter when you pick tooling: does it show the underlying answers, or just a number? We compared the field in our 2026 tools roundup.
The same honesty applies to your own reporting. A single monthly run is one sample; when a movement is within noise, say so. "One month of data; treat it as directional" costs you nothing in the meeting and buys you credibility for the quarter.
How do you run the first client audit in an afternoon?
Enter the client's URL at answermonk.ai. The audit is free, needs no signup, and returns in 3-8 minutes: Three to eight minutes later — no signup — you have the three artifacts from the audit tier above: an appearance rate, a competitor leaderboard, and a citation-source breakdown, built from live buyer-prompt runs on ChatGPT across three search modes..
The audit is not the deliverable. Reading it with the client is. Three things to look for on that call:
- Leaderboard surprises. Incumbency does not carry over into AI answers. our favorite receipt is a Dubai fintech startup showing up in five of every six corporate-card answers (Pemo, 83%), ahead of Emirates NBD and the other major banks. Show a client the upstart outranking the giant in their own category and the meeting changes tone.
- Where the citations point. In our 200-probe Dubai dental set, engines averaged 11.8 citations per answer, and the most-cited sources were the clinics' own websites, not health portals. When owned pages win citations in a category, most of the fix list sits under the client's direct control — an easy first engagement.
- The memory share. If a large slice of the category's answers cite no sources at all, set expectations early: part of this fight is long-term brand mentions, not next week's page edits.
Bring receipts from outside the client's category too. Our public before/after reports live at /reports, and if you are running this across a book of clients, our agency guide covers the multi-client workflow.
What should agencies avoid when selling AI visibility?
Three things kill the offer: guaranteed placements, invented numbers, and astroturfing.
Do not guarantee rankings. With engine presets agreeing only 22% of the time, "we will make you ChatGPT's number-one recommendation" is a promise nobody can keep, and a client who bought a guarantee churns the first time they check a prompt themselves. Sell process and measurement, not outcomes you do not control.
Never invent statistics. The bar in this niche is embarrassingly low — in a July 2026 live probe, one engine's answer about tools to track AI recommendations included a tool whose only web presence was a vercel.app preview URL. That vacuum is exactly why fabricated numbers are tempting, and exactly why real measurement stands out. Clients will eventually verify; build for that day.
And skip the astroturfing. Reddit accounts for roughly 20-24% of Perplexity's citations by industry citation analyses, which makes fake threads look like a shortcut. It is a shortcut to a ban, and the reputational damage lands on your agency as much as your client. The legitimate version — being genuinely useful in communities where the client has standing — is slower, and it works.
The window here is real. Nobody owns the engines' answers in this niche — our July 2026 probes of AI-visibility questions found citations going to small personal blogs, never to vendors. The agencies building proof-first offers now will have a year of receipts by the time clients start asking harder questions — and receipts are the one asset in this niche nobody can fake.
Run your free AI visibility audit →
A correction on llms.txt: earlier versions of this article listed llms.txt as an on-page lever, on the basis that it appeared disproportionately on citation-winning pages in our own crawl. That is a correlation, not a cause. Log-based and controlled evidence since — Ahrefs across 137,210 domains found 97% of llms.txt files were never fetched by anything, and Google states no AI system uses them — indicates the file itself does nothing. We removed the claim. Structure the model can actually read (clear headings, FAQ sections, tables) is the part that holds up.
Where these numbers come from: first-party figures in this article come from AnswerMonk's own studies — the July 2026 fan-out capture (24 buyer prompts across ChatGPT, Claude and Gemini; 893 citations, 806 resolvable, 547 titled), the 21 July 2026 calibration (40 probes, 875 citations), the 2,994-probe study across 19 categories, the 300-probe phrasing study, and city-level probes such as the Chicago and Dubai runs. Each is dated and described, with its sample size and limits, on our methodology page. City-level runs are single-city, single-window samples: treat them as indicative of a pattern, not as population estimates.
Research behind this article
- “Don’t Measure Once: Measuring Visibility in AI Search” (arXiv, 2026) — identical prompts return different brand lists run to run, so single-shot visibility checks are close to meaningless.
- “Optimizing Visibility in Generative Engines: A Critical Survey” (arXiv, 2026) — the peer-reviewed map of what is and isn’t evidenced in this field.
- Ahrefs, 75,000-brand analysis — branded web mentions correlate with AI visibility at r = 0.664 against 0.218 for backlinks: mentions beat links roughly three to one.
Frequently asked questions
What does AnswerMonk's free audit include for an agency?
You enter a client's URL at answermonk.ai with no signup and get results in 3-8 minutes. The audit generates the buyer prompts in that category, runs them on ChatGPT across three search modes, and returns an appearance rate, a competitor leaderboard, and a citation-source breakdown. Agencies can run one per client or prospect, and public before/after examples live at answermonk.ai/reports.
Why do AI visibility tools disagree about who ChatGPT recommends?
Because the engines themselves vary from answer to answer. In our 300-probe phrasing study, two engine presets given identical questions matched winners only 22% of the time in our 300-probe study. That variance is a property of the medium rather than a defect of any one tool, which is why credible reports show the underlying prompts and answers instead of a single score.
Does a client need strong Google rankings before ChatGPT will cite them?
No. Industry analyses find roughly 90% of the pages ChatGPT cites rank on Google page 3 or beyond, so domain authority is not the gate. Bing matters more than most agencies assume, though: Seer Interactive puts ChatGPT Search's overlap with Bing's top organic results at about 87%.
Should agencies push SaaS clients toward G2 and Capterra?
For software categories, usually yes. In our July 2026 live probes, G2 and Capterra were recommended in nearly every answer about getting mentioned by ChatGPT. Review-site presence is also one of the few placements an agency can execute directly, which makes it a natural early deliverable in a retainer.
How quickly can new content show up in ChatGPT answers?
Faster than most clients expect, when it happens at all. Semrush's September 2025 longitudinal test of 81 new pages found 42% were cited by ChatGPT within 30 days, and Google AI Mode cited 36% on day one. The catch is that the remaining pages mostly never got cited, so monitoring matters as much as publishing.
How should an agency price AI visibility work compared to Google SEO retainers?
There is no reliable published benchmark yet, so anchor each tier to work the client already buys: the one-off audit prices like a technical SEO audit, monitoring like a rank-tracking engagement, and the full GEO retainer alongside content or digital-PR retainers. Because tooling like AnswerMonk's free audit costs nothing to run, the entry tier is mostly analysis time, which keeps margins healthy at the door-opener stage.