How can I monitor my search performance across AI-powered search engines? Our setup
To monitor search performance across AI-powered search engines we combine webmaster data, analytics and a visibility platform. This is the setup we use for clients.
"How are we doing in AI search?" is now a standard question in client calls, and it deserves a better answer than "we checked ChatGPT this morning." Here is how we monitor performance across AI-powered search engines on client programs, and where we have learned to be careful.
Split the question in two
Monitoring AI search performance means two different things, and clients often mix them up.
The first is visibility: are we named and cited when buyers ask AI engines about our category? The second is outcome: do any of those answers bring visitors who convert? Visibility data comes from sampling answers. Outcome data comes from analytics. You need both, and they do not come from the same tool unless you pick carefully.
The free layer we always set up first
Before any paid platform, we make sure three free sources are wired up.
Google Search Console. Google reports clicks and impressions from its AI features inside the regular web search data. That makes it useful as a change signal, but we never present it to a client as an "AI Overviews report" because it is blended with classic results.
Bing Webmaster Tools. Copilot draws on Bing, so an unverified Bing property is a blind spot. We set it up on every engagement.
GA4. We build a segment for AI referrers (chatgpt.com, perplexity.ai, gemini.google.com, claude.ai, copilot.microsoft.com) and track engagement and conversions. We tell clients this number is a floor, because some assistants do not pass a referrer.
None of these show whether the brand is named in an answer. For that we need prompt sampling.
Where we run the prompt sampling
Our client programs run on Promptwatch. We agree a prompt set with the client, grouped by topic and intent, and the platform runs it daily per model, country and language. It uses the real interfaces for the live-search platforms (ChatGPT Search, Google AI Overviews, Google AI Mode, Copilot, Gemini, Perplexity, Amazon Alexa, Meta AI) and API access for models such as Claude, DeepSeek, Mistral and Grok. When we report, we say which engines are sampled through which route.
The numbers we put in front of clients are Visibility Score, Share of Voice, Position in Answer and Sentiment, shown as trends rather than single readings. We use the heatmap to compare brands against models or topics against countries, because the pattern usually shows up there before it shows up anywhere else.
Why we also watch crawlers
A big share of "we are not showing up" cases turn out to be access problems. Promptwatch's Agent Analytics shows IP-verified AI crawler logs for bots like GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot, with error details per request. If a page returns errors to a bot, we fix that before we touch the content.
A practical scoping note: crawler logs start on the Professional plan ($245 a month). Essential ($95) does not include them. We flag this early with smaller clients.
Bringing it into one view
The reporting problem with AI search is that the data lives in four places. Two Promptwatch features cut down the stitching for us:
- Search Insights overlays Google Search Console and Bing Webmaster data (including Copilot traffic) on AI crawls and citations.
- 360 Insights joins crawler hits, citation clicks and AI-referred visits per page, so we can say "this page was crawled, cited and visited" or "this page is cited but gets no visits."
That second view is the one clients remember, because it ties monitoring to something they can act on.
Our cadence
- Weekly: a look at top-prompt visibility, crawler errors and the Action Board. The Discovery agent proposes up to five evidence-backed actions per run, and we edit them before anyone acts.
- Monthly: AI-referred traffic and conversions against last month, citation changes by page, and competitor movement. We deliver this as a PDF or a Looker Studio view.
- Quarterly: a prompt-set review, adding prompts when products or markets change.
We never change the prompt set in the middle of a reporting period. If you do, you cannot say whether the brand moved or the questions did.
What we are honest about
Monitoring has limits, and clients are better off hearing them from us first:
- Answers vary between runs, so single readings mean little.
- No tool can guarantee an engine will name a brand.
- Some AI visits never appear in analytics because the referrer is stripped.
- Google's AI feature clicks cannot be separated cleanly in Search Console.
We still think the stack is worth it. It shows where a brand stands, which engines are weak, and whether the work is paying off.
If you are doing this yourself
Set up Search Console, Bing Webmaster Tools and GA4 segments this week. Write 30 to 50 buyer prompts. Check your robots.txt against the real AI bot names. Then try a platform on your prompt set and see whether you miss the extra layers. In our experience, the crawler logs and traffic attribution are what teams end up valuing most.