Guide for agencies
An agency sells AI search visibility by measuring how a client's brand appears in answers from ChatGPT, Perplexity, Google Gemini and Google AI Overviews, then doing the off-site work that changes those answers: publishing content that answers the specific questions being lost, correcting entity records, and earning mentions on the third-party pages the engines already quote. It is sold as a monthly retainer per client, usually alongside existing SEO.
Clients are asking their agencies "what are we doing about ChatGPT?" and most agencies cannot yet answer with work. The behaviour shift behind the question is real: BrightLocal's 2026 consumer survey found 45% of consumers now use ChatGPT or similar tools for local business recommendations, up from 6% a year earlier. Google's AI Overviews reach roughly two billion users a month and appear on a quarter to a third of informational searches.
The commercial consequence is concentration. A traditional search results page lists ten options; an AI answer names three or four. Being one of the named brands is worth more than a mid-page ranking ever was, and being absent is more costly than ranking fifth.
This is the part most agencies get wrong, because the intuition from SEO points at the client's own website.
Across a full audit we ran on a global consumer brand — 232 answers from three engines, every cited link logged — 92% of the sources the engines cited were third-party pages, not the brand's own site. By engine: Gemini 100% third-party, Perplexity 98%, ChatGPT 80%. In the same run, 157 of the 196 answers that named the brand did so without citing the brand's own website at all.
The practical reading: the client's website is roughly 8% of the lever. The other 92% is roundups, comparisons, directories, community threads, editorial coverage and reference records. An AI-visibility service that only edits the client's site is selling a small fraction of the work.
| Work | What it is | Deliverable |
|---|---|---|
| Citable content | Pages written to answer the specific questions the brand loses, better than what is cited today | Published-ready pages, structured for extraction |
| Unlinked mentions | Pages that name the brand without citing it — the most common gap by volume | Ranked target list plus a request to each publisher |
| Third-party mentions | Inclusion in the roundups and comparisons the engines already quote | Secured mention, priced on outcome |
| Entity records | Wikidata, Crunchbase, LinkedIn, industry directories, on-site schema | Corrected, consistent records plus an entity map |
| Technical citability | Crawler access, llms.txt, structured data, content readable without JavaScript | Files and markup, handed over or shipped as a pull request |
Two categories deserve a warning rather than a pitch. Community presence — Reddit, forums, Q&A — is genuinely cited by engines, but the only defensible version is a real, identified expert from the client's team participating honestly. Creating accounts to pose as customers is astroturfing; it breaks platform rules and it is the practice consumer regulators are actively pursuing. Paid placement works mechanically but is poor value: sponsored and advertorial content accounts for roughly 0.3% of AI citations.
They differ in reach, in what they cite, and — critically — in whether they answer at all.
Language matters as much as location. A stiff, translated question often returns no AI Overview at all, while the same question phrased the way a native speaker would ask it returns a full answer. Measuring a Portuguese or Arabic market with translated English prompts produces numbers that describe nothing.
The structure that works is wholesale-in, retail-out, per client slot.
| Layer | Typical range | Note |
|---|---|---|
| One-off audit | Buy ~$1,200 · resell $2,500–3,500 | The sales instrument. Delivers the prompt set, the gaps and the plan. |
| Tracking + reporting | Buy $450–600/mo · resell $1,200–1,800/mo | Measurement and a monthly branded report. |
| Tracking + execution | Buy ~$2,200/mo · resell $4,000–6,000/mo | The work: content, mentions, entity and technical fixes. |
Direct-to-brand providers in this category publish $6,000–15,000 per month for equivalent done-for-you work, which is the ceiling an agency is reselling beneath. Engagements are normally a three-month initial term, then month to month.
Across 232 real engine answers we logged in one audit, not a single answer contained a contact address, and the links that appear are mostly citation machinery rather than the engine sending the user anywhere. The engine hands the user a name, not a path.
That has a commercial consequence most agencies discover too late. The user hears three or four names, then acts — usually by searching the brand directly. The value shows up as branded search, direct visits and shortlist inclusion, not as referral traffic from ChatGPT. A client who measures only AI referral traffic will conclude the work did nothing while their branded search volume climbs.
So agree the scoreboard in writing before the first invoice: named-rate and position on a fixed prompt set, plus the client's own branded-search volume and direct traffic. Report referral traffic if it exists, and call it a floor rather than the measure. A client who learns this at month three has been mis-sold, even if every number they were given was true.
Repeated identical prompts return different brands 40–60% of the time from month to month. Everything below follows from that single fact.
It overlaps, but the weighting is different. Classic SEO optimises the client's site to rank; AI visibility is mostly about which third-party sources exist and what they say, because that is what the engines quote. Technical work on the client's site matters, but it is the smaller share.
No. Any provider offering guaranteed placement in AI answers is either misunderstanding the mechanism or misrepresenting it. Content and mentions raise the probability of being cited; nothing controls the output.
Clients in categories where the AI answer is still contested — mid-market, regional and niche brands with a high-value, researched purchase. Global household names are usually named already and have little to gain. A brand that is invisible in a category the engines answer confidently is the best case.
That is what white-label providers exist for: the agency sells and owns the client relationship, the provider measures and executes under the agency's brand.