What should an AI search analytics setup report so you can defend the budget to leadership
How we build AI search reports that survive a budget review: a fixed prompt panel, citations, crawler access, AI-referred traffic and conversions, in one page.
When a client's finance lead asks why they are paying for AI search work, the worst answer is a screenshot of ChatGPT mentioning the brand. It proves nothing and it cannot be repeated. The reports that hold up in our budget reviews are boring, consistent and tied to money.
Here is the structure we use, and the Promptwatch features it runs on.
Start with a prompt panel you never change
We agree a panel of buyer prompts with the client, grouped by topic and by intent: category questions, comparison questions, brand questions. Then we freeze it. Additions go in a separate group so the trend line stays honest.
On that panel we report Visibility Score, Share of Voice against the named competitors, and Position in Answer. Promptwatch runs the panel daily per model, country and language, so you get a time series rather than a snapshot. Sentiment gets a weekly summary, which is useful context but rarely what a CFO asks about.
Be careful with personas. The same prompt asked as a procurement manager and as a founder can return different brands. Persona stacking in Promptwatch lets you track both without mixing them in one number.
Explain visibility with citations
Visibility is a result. Leadership wants the cause. Citation analytics shows which pages and domains get cited for your prompts, including Reddit, YouTube and LinkedIn reports. In practice this is where the budget argument becomes concrete: "Three competitor comparison pages account for most citations on our commercial prompts, and we have nothing equivalent." That sentence funds a content plan.
The content gap analysis and the Action Board then turn gaps into a ranked to-do list. We show leadership the list with owners and dates, because a report that ends in actions reads as a program, not an experiment.
Show whether the crawlers can reach the site
This is the section most agencies skip, and it is where we find the fastest wins. Agent Analytics reads real AI crawler logs, IP-verified so spoofed user agents are filtered out. We watch GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot and PerplexityBot, then look at errors per page and at which crawled pages actually get cited.
A client with blocked or erroring pages should fix that before buying more content. Saying so in a budget review earns trust, because it shows we are willing to recommend spending less. Note that crawler logs begin on the Professional plan at $245 a month, so we flag it early when a client starts on Essential.
Connect AI visits to conversions
Visitor analytics counts AI-referred visits from ChatGPT, Perplexity, Gemini, Claude, Copilot and others, with conversions, using a lightweight script, the GTM template or the WordPress plugin. We report AI sessions, conversion rate and conversions next to organic and paid, with the same definitions.
Keep expectations calibrated. For many sites the volume is still small. We say that in the first sentence of the section and show the trend, not just the level. Google's AI features are harder to separate from classic Search traffic, so the Search Insights view that overlays Google Search Console and Bing Webmaster data helps, but we label estimated numbers as estimated.
Promptwatch's published case studies show how this can play out for customers, for example Crisp reporting AI-referred traffic converting at twice the rate of traditional channels. We cite it as their result and never as a forecast for a client.
Package it so it gets read
Our monthly pack is one page plus an appendix:
- Visibility on the fixed panel, versus the previous period.
- The three largest citation gaps and what we did about them.
- Crawler health: errors found, errors fixed.
- AI-referred sessions and conversions.
- Next month's actions and their cost.
Promptwatch generates scheduled PDF reports from any dashboard, and custom reports with the Looker Studio connector are available from Professional, so the pack builds itself and we spend our time on the commentary. Agent Chat can draft a first read of the numbers, which we always check before it goes out.
What we tell leadership we cannot do
We cannot guarantee that any engine recommends a brand. Models change, and answers vary run to run. What we can guarantee is measurement: a stable panel, a clear view of why the numbers move, and a list of actions with owners. We say this up front because it makes every later number more believable.
A quick test of your own setup
Ask whether your current reporting can answer these in under a minute: which pages get cited, which AI bots fail on your site, how many conversions came from AI sources last month. If any answer is "we would have to check", that is the gap to close first, and it is usually cheaper than another round of content.