Clients are starting to ask why ChatGPT never mentions them. Usually they heard it from a prospect, or they asked an assistant themselves and got a competitor's name. That question is the opening for a new service line, and it is a service line most agencies can run without hiring anyone new.
What follows is a packaging playbook: what goes in the audit, what goes in the monthly report, how to price it, what tooling you actually need, and the promises you should refuse to make even when a client pushes.
Start With the Audit, Not the Retainer
Sell a one-time audit before you sell a subscription. It gives the client a concrete artifact, gives you a baseline to measure against, and filters out the people who want a report but not the work.
A defensible audit has four parts.
1. Question inventory
Write down the 20 to 40 prompts a real buyer would type. Not keywords — full questions. "Best commercial litigation firm in Dallas for partnership disputes" beats "Dallas lawyer." Pull from sales call recordings, your existing keyword research, and the client's own FAQ page.
Group them by intent: category discovery ("what software does X"), comparison ("X vs Y"), problem-first ("how do I fix Z"), and branded ("is X any good"). Category and comparison prompts are where visibility is won or lost.
2. Baseline across assistants
Run each prompt in ChatGPT, Gemini, and Perplexity and record four things per prompt:
- Whether the brand is mentioned at all
- Whether it is linked or cited
- Which competitors appear
- What source the assistant used (the client's site, a directory, a review site, a news article)
Save screenshots. Assistants are non-deterministic — the same prompt can produce different answers on different days, and sometimes on the same day. Screenshots are how you show a client what you saw and when.
3. Crawl and access check
This is the part clients cannot do themselves and the part that most often explains a total absence from answers. Check robots.txt against the documented crawlers.
| Crawler | What it does | What blocking it affects |
|---|---|---|
| OAI-SearchBot (OpenAI) | Surfaces sites in ChatGPT search | Sites opted out are not shown in ChatGPT search answers |
| GPTBot (OpenAI) | Collects content to improve foundation models | Does not affect ChatGPT search |
| ChatGPT-User (OpenAI) | User-initiated actions in ChatGPT and custom GPTs | User-initiated, so robots.txt may not apply; not used for search inclusion |
| ClaudeBot (Anthropic) | Content that may be used for training | Excludes future content from training |
| Claude-User (Anthropic) | Fetches pages when a user asks Claude a question | Prevents retrieval for user queries |
| Claude-SearchBot (Anthropic) | Indexes content for search results | Prevents indexing for search |
| PerplexityBot | Surfaces and links sites in Perplexity results | Not used to crawl for foundation models |
| Perplexity-User | Fetches pages for user questions | Generally ignores robots.txt because the user requested it |
| Google-Extended | Training for future Gemini models; grounding in Gemini apps and Vertex AI | Does not affect Google Search inclusion and is not a ranking signal |
Two things to say out loud in the audit. First, Google AI Overviews are part of Google Search, so they are governed by normal Googlebot crawling and indexing — Google-Extended is not the lever there. Second, robots.txt expresses intent; it is not access control, and changes can take about 24 hours to propagate for OpenAI and Perplexity.
A surprising number of clients have blocked a search-related crawler while trying to block a training crawler, or the reverse. That single finding often justifies the audit fee.
4. Source and content gap
For every prompt where a competitor appears and the client does not, note the source the assistant leaned on. Then answer one question: can the client realistically be on that source? If the answer is a third-party review site, that is a different workstream than adding an FAQ section. Be honest about which gaps are addressable and which are not.
What the Monthly Report Should Contain
Keep it to two pages. Clients read the first page.
Prompt coverage table. Same prompts, month over month, with mention rate and citation rate. Show movement, including downward movement.
Share of voice. How often the client appears versus a named competitor set. Pick three to five competitors and stick with them so the trend line means something.
Source shifts. Which domains the assistants cited this month. This is the most actionable section, because it tells the client where to publish or get listed next.
Work log. What you actually did: pages published, schema added, listings claimed, crawler settings changed.
Next month's plan. Three items, no more.
A note on measurement noise
Assistants do not return stable answers. If you report a single number with no context, a client will eventually run the same prompt, get a different answer, and conclude you fabricated the report. Handle this by reporting rates over a fixed prompt set rather than individual answers, and by stating plainly that answers vary. "Mentioned in 9 of 30 tracked prompts this month, up from 6" is defensible. "You now rank #1 in ChatGPT" is not.
What to Promise and What to Refuse
Promise:
- A documented baseline and repeatable monthly measurement
- Crawler and indexing configuration that matches the client's intent
- Content and source work aimed at the specific prompts where they are absent
- Clear reporting on what changed and what did not
Refuse, every time:
- A guaranteed mention, citation, or position in any assistant
- A specific timeline for appearing in answers
- The ability to control what an assistant says about the brand
- Any claim that a single tactic will produce visibility
The honest framing: you are improving the inputs assistants draw on and removing technical barriers to being included. You are not buying placement, because there is no placement to buy.
Pricing
Three common structures, with the tradeoffs.
Flat audit fee, then retainer. Simplest to sell. The audit should be priced so it is profitable on its own, not as a loss leader — some clients will take the audit and stop. A useful test: if you would resent doing the audit at that price for a client who never renews, it is priced too low.
Audit bundled into a three-month minimum. Lower entry price, better retention, but you carry the risk that the client's category is genuinely hard to enter. Only offer this when the audit shows addressable gaps.
Add-on to an existing SEO or content retainer. Easiest upsell, lowest friction, but it tends to get deprioritized when the core retainer gets busy. If you go this route, give it its own line item and its own monthly deliverable so it does not disappear.
On scope: cap the tracked prompt set. Thirty prompts is a real report. Three hundred is a data dump nobody reads, and it multiplies your QA time without changing the recommendations.
Tooling
You need three capabilities: prompt running across assistants, citation capture, and historical tracking.
You can start manually with a spreadsheet and screenshots. It works, it is slow, and it does not scale past a handful of clients. When you outgrow it, a purpose-built tracker is worth the cost — PromptRanking, for example, is built to track whether ChatGPT, Gemini, and Perplexity recommend a brand. Whatever you choose, check that it stores history, because the month-over-month comparison is the product.
Separately, keep a crawler and log audit in your stack. Server log analysis tells you which AI crawlers are actually hitting the site, which is often different from what robots.txt says should be allowed. If your clients are local or in a specific vertical, the mechanics differ enough to be worth reading up on — see our notes on local businesses and AI assistants, dental practices, and law firms.
The First Step
Pick one existing client, run the 30-prompt baseline by hand this week, and check their robots.txt against the crawler table above. Bring the findings to your next call as a free sample. If the client reacts the way most do when they see a competitor named and themselves absent, you have your first paid audit — and a template for the next twenty.