How do I show a client which competitors AI assistants recommend over them?

Quick Answer: Log which competitors AI assistants recommend for ten buyer prompts, note cited sources, and report the share-of-voice gap. If a competitor is named more often, the gap is real. At similar rates, track cited sources. This affects any agency or brand whose buyers ask AI assistants for recommendations, and the sources give no reliable count of how many.

Key takeaways

• Write every panel prompt without the client's brand name, then add a second phrasing for the questions that matter most.
• Keep the prompt list frozen between runs, and version it when it has to change.
• Put branded search clicks next to AI referrals in every client report.

How do I show a client which competitors AI assistants recommend over them?

Show the client evidence they can rerun themselves. Build a fixed panel of buyer questions, log every brand each assistant names and every source it cites, and turn the log into one share-of-voice number per competitor.

Start by hand, on one assistant:

1. Write ten questions a buyer asks before choosing a supplier in the client's category, with no brand names in them.
2. Open the assistant, start a new chat for each prompt, and paste the prompt exactly as written.
3. In a sheet, log one row per answer: prompt, assistant, date, country, language, every brand named in the order it appears, and every cited URL.
4. Add a column per brand that counts whether it was named in that answer, so each prompt shows a competitor count.

Expected result: every row has a named-brand list and at least one cited source, so you can tally a competitor count per prompt.

Which prompts and assistants belong in the panel?

The panel should hold the questions buyers ask before they choose, run on every assistant those buyers use: ChatGPT, Gemini, AI Overviews, AI Mode, Claude and Perplexity. Google describes the relationship this way: "AI Mode expands what AI Overviews can do with more advanced reasoning and ways of interacting" [1]. Merging the two into one "Google AI" row hides which surface the client loses on, so log them separately.

The common mistake is testing branded queries only. One practitioner guide on the byaxovox blog argues that an AI answer can put a competitor first even for a branded query, because "The AI isn’t loyal to the query, it’s loyal to the evidence" [4]. Build the panel from non-branded buyer questions.

Google's help page for AI Overviews tells searchers to "Ask multiple versions of your question to get the best answers" [2]. Your client's buyers will do the same, so a second phrasing of the key questions belongs in the panel. AI Mode sometimes shows a set of web links instead of an AI response when Google's confidence in the answer is low [1]. Log both cases as "no AI answer" instead of deleting the row, so the denominator stays honest.

Record country and language on every row.

Once the log fills, each pattern points to one action:

Why an assistant names a competitor instead of your client

An assistant usually names a competitor because the competitor's public evidence passed a check that the client's did not. The byaxovox guide puts it this way: "Those systems check a business before naming it, and a business that fails the check doesn’t get an error message or a warning letter" [4]. The client is simply replaced. The same author lists seven criteria: retrievable, identifiable, quotable, specific, corroborated, current and associated [4]. Treat that as one practitioner's working list, not a rule any platform has published.

Google's own description of AI Mode explains why page-level thinking misleads: "It divides your question into subtopics and searches for each one simultaneously" [1]. The practitioner reading is that the answer is assembled from the best passage for each hidden sub-query [4]. Compare passages instead. Write out the likely sub-questions behind a lost prompt, put the client's best paragraph for each next to the competitor passage the assistant cited, and judge which one answers more directly and specifically.

Can getting mentioned in ChatGPT actually bring new customers?

It can, but referral traffic shows only part of the effect, so never promise an agency's e-commerce client a number or an ROI figure. In one practitioner's view, the byaxovox guide argues that many AI-recommended buyers take the brand name, search it on Google, and arrive as branded organic search: "The chatgpt.com line in your analytics is a floor, not a total" [4]. Seer Interactive adds that "Not all AI platforms expose their identity when sending traffic", and that privacy-sensitive queries may still strip referrer information [5].

Reporting chatgpt.com referrals as the full return is the trap. Label that line as a floor and put the branded search trend beside it:

1. In GA4, pull weekly sessions with the source chatgpt.com for the last eight weeks.
2. In Search Console's Performance report, filter queries that contain the client's brand name and pull weekly clicks for the same eight weeks.
3. Chart both series on one axis of weeks, and add the dates of panel runs where the client's share of voice changed.

Expected result: a referral line and a branded-search trend the client can read together, with the referral line labelled as a floor.

What the client report should show

The report should show five measures per brand: visibility, share of voice, position, sentiment and cited sources. Welt Analytics's example dashboard lists the same set [3]. Rerun identical prompts on the same assistants, country and language at a fixed cadence.

Worked example

Say a client's panel has 10 prompts run on 3 assistants, giving 30 answers. The client is named in 6 of 30 answers, 20%. Competitor A is named in 15 of 30 answers, 50%. Competitor B is named in 6 of 30 answers, also 20%. Against Competitor A the gap is 30 percentage points, which is real and goes in the report with the cited sources behind it. Against Competitor B the rates match, so you track which sources each one wins on instead of reporting a loss.

If you would rather not rebuild the sheet each week, Welt Analytics daily AI answer checks run this kind of panel for you. Its homepage says: "Every day we check how ChatGPT, Gemini, AI Overviews and AI Mode answer what your buyers ask (Claude and Perplexity weekly), and where you sit next to competitors." [3]

How do I optimize my Shopify store for AI search engines like ChatGPT and Perplexity?

Start from the panel's cited-sources column. It shows which pages and third-party sources won each answer, and so which of the client's pages to fix first.

Treat llms.txt as unproven. The byaxovox guide reports that Google says on record it does not use llms.txt [4], an unverified claim from that blog.

If you want this panel run for you, with daily checks on four assistants, weekly checks on two more and competitor positions logged, see Welt Analytics.