How to Judge Generative Search Optimization Case Studies

A trustworthy generative search optimization case study names what was measured, how, over what period, and against what starting point, and it admits what the work could not control. Many case studies in this young field show a big percentage and a screenshot. Knowing what to look for helps you separate real evidence from marketing.

Why Case Studies in This Field Are Tricky

AI answers change from one run to the next, models are updated without notice, and assistants use different sources. A screenshot of ChatGPT mentioning a brand proves only that it happened once. Traffic from AI tools is also hard to attribute, because many visits arrive without clear referral data. None of this makes case studies useless, but it means method matters more than headlines.

What a Solid Case Study Includes

• A defined set of prompts, ideally listed, chosen before the work started.
• Repeated measurement: each prompt run several times per assistant, so variation is accounted for.
• A baseline recorded before any changes.
• A clear time frame, with dates.
• The actual work done: what content, where, how much.
• Other factors noted, such as a product launch, press coverage or seasonal demand that may have contributed.
• Business outcomes, where available, measured honestly: enquiries, sign-ups, or “found us via AI” responses.

Red Flags

• Percentages with no base: “300% increase in AI mentions” might mean going from one mention to four.
• Single screenshots presented as proof.
• Anonymous clients with no detail at all, combined with very large claims.
• Claims that results were “guaranteed” or achieved through special access to AI companies.
• No mention of what did not work.

Questions to Ask a Provider

1. Which prompts did you track, and who chose them?
2. How many times was each prompt run, and in which assistants?
3. What was the starting point?
4. What else changed for the client during that period?
5. Can I speak to the client, or see a redacted report?
6. What results did other clients in similar situations get, including weaker ones?

An Illustrative Example (Not a Real Client)

Imagine a regional bicycle repair chain tracks 20 prompts like “bike repair shop open Sunday in [city]” across five assistants, three runs each, for a baseline of 300 answers. It appears in 30 of them. Over four months it fixes inconsistent hours on directories, publishes service pages for each location and earns a local news feature. It then appears in 75 of 300 answers. A good write-up would report both numbers, note the news feature as a possible factor, and list the prompts that did not move. That honesty is what makes it believable.

Running Your Own Mini Case Study

The most reliable evidence is your own. Choose your prompts, record a baseline, do the work, measure again with the same method. Our page on measuring AI visibility explains how, and AI search optimization case studies covers what types of results are realistic.

How Keboost Helps

Keboost does not publish invented success stories. Instead, it gives you the tools to measure your own results: it tracks a fixed set of prompts across ChatGPT, Gemini, Claude, Perplexity and Copilot over time, so you have a baseline and a trend. It also creates and distributes focused content to improve your chances. Run a free AI visibility check to record your starting point.