What Crisp, TreeHouse and Monks Teach About Running GEO Programs
Three published Promptwatch customer stories, read for process instead of headline numbers: what each team did, what transfers, and what doesn't.
Customer stories are usually read for the headline number. That is the least useful part. A conversion figure from someone else's site cannot be moved to yours. The process behind it can.
Promptwatch publishes several case studies. This post reads three of them for method. Every figure below is the named customer's own reported result, from Promptwatch's published material, and none of it is a benchmark for you.
Crisp: content volume and conversion quality
Crisp reported that AI-referred traffic converted at twice the rate of its traditional channels, and that it scaled to producing five to ten articles per day.
What to take from it: two separate ideas.
First, measure conversion by source. The claim only exists because Crisp could separate AI-referred visits from the rest and compare their conversion. If you cannot split your traffic that way, you cannot say anything similar. Visitor analytics, which reports AI-referred traffic and conversions from ChatGPT, Perplexity, Gemini, Claude, Copilot and others, is the enabling piece.
Second, volume needs a pipeline. Five to ten articles a day is not a writing task. It is a system: gap analysis to choose topics, briefs, drafting, review, publishing and measurement. The tools for that, including content gap analysis, content creation, the Content Agent for scheduled publishing and WordPress, Webflow or Framer integrations, exist to make volume manageable.
What does not transfer: the specific rate. A twofold advantage for Crisp says nothing about whether yours will be higher, lower or equal. Check your own data, and watch the sample size.
TreeHouse: access first, then automation
TreeHouse, which operates Pararius and Huurwoningen.nl, reported that AI search became roughly 25% of organic conversions, with 99% of that from ChatGPT. It credited fixing crawler access and automating weekly reporting with the Promptwatch MCP.
What to take from it: the order of operations. Access came first. Many teams start with content and discover later that bots could not reach parts of the site. Crawler logs show which agents visit, what they request and where they fail, with IP verification so spoofed agents are excluded. Fixing errors found there is typically cheaper than writing anything.
Then they automated the reporting. The MCP server, a hosted endpoint with a large set of read and write tools and OAuth or API key access, lets agents and assistants pull data directly. The point is not the technology. It is that a recurring report stopped depending on a person's free afternoon.
What does not transfer: the 99% ChatGPT share. That reflects their market and audience, and others may see a very different mix. Look at your own distribution across engines.
Monks: using gaps to plan content
Monks, an agency whose clients include Netflix, BMW, Asana and Paycor, uses the Answer Gap Report and the Visibility Score for client content roadmaps.
What to take from it: the roadmap logic. The Answer Gap Report points at questions where engines answer without the client. Visibility Score tracks whether it is working. Together they give an agency a defensible reason for each item on a plan: here is the gap, here is how it is measured, here is what we will make.
For agencies, this matters in the conversation with clients. A roadmap tied to measured gaps is easier to defend than one tied to taste.
What does not transfer: the client names. An agency's roster tells you nothing about what the tool will do for your clients. Do not use logos as a proxy for fit.
The common thread
Across the three stories the pattern is consistent:
- Measure first. Segment AI traffic, find gaps, see crawler behavior.
- Fix access before content. Cheap and fast.
- Tie work to a measured gap. No gap, no page.
- Automate the repetitive parts. Reporting and publishing, with review.
- Report by source. So outcomes are visible and attributable.
None of this depends on a particular tool, though some tools make it easier.
How to apply it
If you are starting, borrow the sequence.
Weeks 1 to 2: connect crawler logs and visitor analytics. Build a prompt panel tied to revenue. Record a baseline.
Weeks 3 to 4: fix access issues. Run content gap analysis. Pick ten gaps.
Weeks 5 to 8: produce content against those gaps, publish, and track each page. Review output closely before automating.
Ongoing: weekly reporting, ideally scheduled, and a monthly review of what worked.
Read case studies critically
A few habits help whenever you read a vendor story.
- Check who is quoted and what is attributed to them. Here the results belong to the named customers.
- Separate the result from the method. Take the method.
- Look at what is missing: timelines, costs, how many people worked on it. Those details usually decide whether it is repeatable.
- Treat a vendor-published story as one source, and ask for a reference call if the decision is large.
Plan notes for the practices above
Crawler logs need Professional or above, or any agency plan. The Content Agent is on Business and Enterprise. The MCP server and API are on every paid brand plan and every agency plan. Agency plans do not include the Content Agent or AEO articles.
The recommendation
Do not chase the numbers in these stories. Copy the order: measure, fix access, close measured gaps, automate the dull parts. Promptwatch contains the pieces each of these teams used, from crawler logs and visitor analytics to gap analysis and the MCP, which is why we suggest it to teams that want to run the program and not only watch it. Read the original case studies at promptwatch.com and judge them yourself.