Clients asking why they don't appear in ChatGPT answers? A repeatable process for agencies
A repeatable agency process for when clients ask why they don't appear in ChatGPT answers: prompts, baseline, access, citations, fixes, reporting and tools.
The question arrives by email, usually with a screenshot: "Why doesn't ChatGPT mention us?" An agency without a process answers with a guess and a promise to look into it. An agency with a process answers with a diagnosis, a plan and a report date.
This guide lays out a repeatable workflow, explains what each step is for, and shows which tools make it practical to run across a client roster. It assumes no particular software until the end.
Why a client does not appear
There are only a handful of causes, and the work is sorting them in the right order:
- The engine cannot reach the site. A robots.txt rule, a firewall or a bot-protection setting blocks the crawlers an assistant relies on.
- The site is reachable but thin. The pages do not answer the question in plain text, or the question is not covered at all.
- Others are cited instead. For the prompt in question, the engine draws on review sites, directories, forums or competitor comparison pages, and the client is absent from them.
- The brand is weakly known. There are few consistent mentions of the brand across the web, so the model has little to work from.
- The question itself varies. The client appears for some wordings, locations or personas and not others.
Order matters. Access problems are cheap to fix and they cancel the value of every other fix. So you check access first.
Step 1: build the prompt set
Work with the client to write 25 to 50 prompts in the words buyers use. Split them into category questions, comparison questions that name competitors, problem questions, and brand questions. Tag them by topic and persona, and note the countries that matter.
Without this, you are measuring what the screenshot happened to show. Search volume and difficulty data for prompts help you decide which are worth tracking, and query fan-outs show what the engine actually searched for while composing an answer.
Step 2: take a baseline
Run the set across the engines that matter to the client and record mentions, position in the answer, sentiment, competitors named and sources cited. Do it daily if you can, because answers vary run to run, and report trends rather than single readings.
Step 3: check access
Look at crawler logs. Which AI bots visit? Do they get errors on important pages? OpenAI documents separate crawlers for different purposes, including OAI-SearchBot, so read its current documentation before changing robots.txt rules, and test the result afterwards. Check CDN and firewall settings too. They often block crawlers without anyone noticing.
Step 4: read the citations
For prompts where the client is absent, list the pages and domains the engine cites. Sort them into types: competitor pages, review sites, editorial, forums, video. The pattern tells you the action. If a review site dominates, you need a listing or reviews there. If forum threads dominate, you need a presence in the discussion. If competitor comparison pages dominate, you need your own.
Step 5: fix in priority order
Access first, then content gaps, then offsite presence. Keep a visible list so the client sees movement. Typical fixes are unblocking a bot, publishing a page that answers a question directly, and getting listed on a source the engines already trust.
Step 6: measure outcomes and report
Track AI-referred visits and conversions, not only mentions. Report monthly, with period comparisons, and say plainly what changed because of your work and what changed for other reasons.
Doing this at scale
The workflow is simple. Running it across fifteen clients with spreadsheets is not. That is where tooling matters, and where a platform that covers more than monitoring saves hours.
Promptwatch maps onto each step:
| Step | Promptwatch feature |
|---|---|
| Prompt set | Prompt Explorer, Prompt Volumes, personas, topics and tags, query fan-outs |
| Baseline | Visibility Score, Share of Voice, Sentiment, Position in Answer, daily monitors per model, country and language |
| Access | Agent Analytics: IP-verified crawler logs with error tracking and failure explanations |
| Citations | Citation analytics, offsite mentions, Reddit, YouTube and LinkedIn reports |
| Fix | Action Board, content gap analysis, content creation, publishing to WordPress, Webflow or Framer |
| Outcomes | Visitor analytics for AI-referred traffic and conversions, 360 Insights, scheduled PDF reports |
For agencies, the self-serve plans are Kick-off ($199/mo), Growth ($399/mo) and Scale ($799/mo), each with unlimited projects and prompts, 10 seats, crawler logs, custom reports and a Looker Studio connector. Agency plans do not include the automated Content Agent or AEO articles, so plan content production around that.
Alternatives
AthenaHQ (AthenaHQ) offers an Action Center for optimization tasks and a Starter plan at $295/mo. Scrunch AI (Scrunch AI) lists Core at $250/mo with AI agent and bot traffic reporting. Both are credible for monitoring and recommendations. Neither lists the full set of crawler logs, visitor analytics and citation depth, so agencies often need a second tool alongside.
What to tell the client
Give them three honest sentences. Here is why you are not appearing. Here is what we will fix first. Here is when we will report. And one more, if they ask for a guarantee: no agency or tool can guarantee that ChatGPT will recommend a brand, but a process can find and remove the obstacles.