How do I audit a new client's visibility in AI search before a pitch?
Quick Answer: Before a pitch, run 15 to 20 buyer-style prompts through ChatGPT, Gemini, Claude, Perplexity and Google's AI Overviews and AI Mode. Log whether the client is named, its position, cited sources and rival brands. Absent while competitors are cited: pitch content and source work. Named but framed badly: pitch fixing cited sources. Named widely: pitch monitoring.
Key takeaways
- Log a missing AI Overview or a links-only AI Mode result as
no AI answer, and leave it out of the new client visibility gap you pitch. - Trace every wrong claim about the prospect to the page it came from before you pitch a fix.
- Put the word "snapshot" and the run date on every audit slide.
- Quote revenue measurement as a later phase of the proposal.
What does an AI search audit of a new client actually measure?
An AI search audit measures what the engines say about a prospect for a fixed set of buyer questions. Is the client named? Where in the answer? In what tone? Based on which sources? And which competitors get named instead? Those questions map to the six fields Welt Analytics reports for a brand in AI answers: visibility, share of voice, position, sentiment, sources and competitors. Use them as your column headers even if you work by hand.
This affects any agency or consultant pitching a prospect whose buyers use AI assistants to research products, whether or not the prospect has any AI presence yet. A brand with zero mentions still gets a full audit. The empty grid is the finding.
One thing catches people on a first audit: a blank result is not always a miss. Google's help page says "You'll find AI Overviews in your Google Search results when our systems determine that generative AI can be especially helpful" [1]. Some prompts do not trigger one. AI Mode has its own exit. Google's AI Mode page says: "In some cases, AI Mode will provide a set of web links if there’s not high enough confidence in the quality or helpfulness of an AI response." [2]
Those cases belong in the grid as no AI answer. Counting them as the client being left out inflates the gap you pitch.
How do I build the prompt set for a new client visibility audit?
Build the set from 15 to 20 questions your prospect's buyers would ask, split into three groups. That size is a working rule, not a benchmark.
- Category questions are what a buyer asks before they know any brand:
best [category] for [buyer type]orwhich [product] for [use case]. - Comparison questions put the prospect next to a rival:
[prospect] vs [competitor]andalternatives to [competitor]. Pick rivals the prospect names in its own sales material or that already sell in its market. - Branded questions test what the engines know:
is [prospect] any good,[prospect] pricing,what do customers say about [prospect].
Phrase each prompt the way a buyer types into a chat box: a full question with context, such as company size, budget or use case.
Keep the list in a sheet, one row per prompt, so the same set can be re-run after the pitch.
How do I run the audit and read the results?
Run every prompt on every engine, log the same fields each time, then sort the rows into the patterns in the table below. If you later want to track client brands across ChatGPT, Gemini, Claude and Perplexity over months, this grid is your first data point.
- Open a sheet with these columns:
prompt,engine,named(yes or no),position(first brand named, second, and so on),sentiment,cited URLs,competitors. - Run each prompt in ChatGPT, Gemini, Claude and Perplexity. Then run it in Google Search to see whether an AI Overview appears, and in Google AI Mode.
- Fill one row per answer. Copy the cited URLs in full.
- Mark a row
no AI answerwhen Google shows no AI Overview, or when AI Mode returns only links. - Match each row to a pattern in the table below.
Expected result: a grid of prompts by engine. It shows which prompts leave the client out, and which sources and competitors the engines name instead.
The next trap is treating one run as the truth. Welt Analytics's about page puts it plainly: "AI engines don't answer the same question the same way twice." [3] The fix is to run each prompt more than once and report the pattern. (You will be tempted to skip the repeats. Don't.)
To check, re-run three of the prompts you logged, on the same engine. Expected result: the same competitors and sources recur in most of the repeat answers, so the gap is a pattern you can pitch. If the answers differ completely, add more runs before you report it.
Google warns that "AI Overviews can and will make mistakes." [1] When an answer gets the client wrong, open its cited links first; a cited page that carries the error is where the fix starts.
| Pattern in the grid | How to confirm | Pitch action |
|---|---|---|
| Client never named, competitors named | Repeat runs show the same competitors, and the client is still missing | Pitch content that answers those prompts and a baseline to track share of voice |
| Client named but never cited as a source | The cited URLs column holds no page from the client's own domain | Pitch publishing pages the engines can cite on the prompts where the client appears without a link |
| Client named with wrong or negative framing | The cited links include the page the wrong or negative claim comes from | Find the sources cited in those answers and pitch correcting them |
| Client named on most prompts | named reads yes on most rows, across engines and repeat runs | Pitch ongoing monitoring and defending position against named competitors |
Can I do this by hand, or do I need a tool?
By hand works for one prospect and one afternoon. It stops working once the same prompts need checking on a schedule. Twenty prompts on six engines is 120 answers per run, before any repeats.
The trap here is calling a one-day manual audit a trend. It is a snapshot: label it that way on the slide, with the run date.
A tracking tool runs those repeat checks for you. Welt Analytics checks ChatGPT, Gemini, Google AI Overviews and Google AI Mode daily. It checks Claude and Perplexity weekly. It also scans the client's site and sets the first prompts. The Starter plan tracks 50 prompts and 1 brand across all 6 engines for 89 EUR per month, excluding VAT (price as of 2026-11-13).
If you are weighing tools, see how to set up a GEO tool for a new client in one day.
What do I put in the pitch once I have the numbers?
Turn the grid into three slides: where the client stands, who the engines name instead, and which sources shape the answers.
The first slide is the grid itself, with the prompts where the client is absent highlighted. The second lists the competitors that recur across engines. Your guide to show a client which competitors AI assistants recommend covers how to present that contrast. The third groups the cited URLs by site.
Do not promise the prospect they will appear in AI answers. Nobody controls that. Pitch what you control instead: a measured baseline, the content and source work from the table, and a monitoring plan that re-runs the same prompts.
This audit counts mentions, not money; revenue from ChatGPT or Perplexity traffic belongs in a later phase of the proposal. When the client asks how you will show progress month to month, point to how you report AI search visibility to clients.
Running pre-pitch audits for several prospects a month? Track the prompt sets daily and weekly across the engines with Welt Analytics.
Questions
Will I need more prompts once the client signs?
For ongoing work, a bigger set pays off, split by product line, region or buyer type. For scale, Welt Analytics's Growth plan tracks 150 prompts and 2 brands, and Pro tracks 400 prompts and 3 brands.
Should I audit the prospect's competitors too?
Your grid already captures who gets named. Go one step further: run a few branded prompts about the one or two rivals that appear most often. That shows what the engines say about the brands winning the answers.
Do I need an account or a paid tool to start?
A manual audit needs only access to the engines themselves. If you want the repeat runs automated, Welt Analytics offers a 7-day free trial with no card required.