For SEO agencies

AI visibility metrics to report to clients when traffic drops: 4 numbers (2026)

A client's clicks fall and they ask: "Where did the traffic go?" You cannot prove AI took the clicks. You can show one measurable thing: how often AI assistants name the brand when buyers ask, reported as a trend, because one run is a snapshot.

Drafted by our AI research agents. Reviewed and approved by Jordan, founder. Published November 6, 2026. 8 min read.

One run is a snapshot: we asked the same 10 questions about Moz twice on October 29, 2026, a few hours apart, and Moz was named in 14 of 40 answers, then 18. Get these four numbers for one of your clients, free: 48 hours, your logo, no card.

Why report AI visibility when traffic is down?

Because the client is asking where the traffic went, and AI answers may be one cause. A count of how often the brand is named gives the agency something measurable to say, with a date and a method.

AgencyAnalytics, which sells an AI tracker, makes the same case in a sponsored Search Engine Journal article (August 11, 2026); we could not verify its survey figures. AI visibility is not a conversion: put the count next to the client's conversion data.

Why report a trend, not one snapshot?

Because the same question, asked again, does not always give the same answer. A gain of a few answers may be noise, and you do not want to present it as your work.

We ran the same 10 questions about Moz twice on October 29, 2026, a few hours apart, on four assistants (40 answers per pass). Same 10 questions, same four models, same API options. Pass 1 was collected over 22 minutes; pass 2 was one run of about 2 minutes.

What moved. Moz +4, Semrush +3, Ahrefs +1, SE Ranking -2. Per assistant, Moz went from 2 to 2 (ChatGPT), 7 to 9 (Claude), 2 to 4 (Gemini) and 3 to 3 (Perplexity). The first tracked brand changed, or none was named, in 7 of the 40 answers (Gemini 4 of 10, the others 1 each).

What held. In 146 of 160 yes/no checks ("is this brand named in this answer?") both passes agreed (102 times the brand was named in both, 44 times in neither), and "best tool to check domain authority" had Moz first on all 4 assistants both times.

The limits. Two passes, the same day, one answer per question per pass, official APIs with web search on. Moz is a well-known brand; a smaller one may behave differently. This shows that answers move, not by how much, so we give no threshold. The method it supports: fixed questions, more than one run, the gap between runs as your baseline, and "no clear change" when a monthly move is within it. The free sample shows your client's counts, sentences and cited pages from one run; the full report asks each question 3 times, so you can read a monthly change against repeated runs. If a client's own check disagrees with these counts, here is what to say.

What are the 4 AI visibility metrics to report?

Each is a count a client can check, from the same buyer questions asked of every assistant.

Pass 1 is the one in our 3-page Moz sample report. Place is the rank among the four brands tracked (1 = named first); a question counts as won when at least 3 of 4 assistants agree, so Moz is first on 1 question but first in 4 single answers.

• 1. Share of answers. Moz was named in 14 of 40 answers: ChatGPT 2 of 10, Claude 7, Gemini 2, Perplexity 3. Show it per assistant, because one total hides the spread.
• 2. Place. Place 1 means named first among the four brands tracked, not among every tool the assistant mentions. Moz came first on 1 of 10 questions, "best tool to check domain authority". On 6, a competitor came first on at least 3 of 4 assistants; on 3, the assistants split and the report says "Varies". The same split held in pass 2.
• 3. Cited pages. The report lists the full address of each page, not just the site. Cited most often: semrush.com (12 of 40 answers), onelittleweb.com (9), ahrefs.com (7); Moz's own site, 3. One full address, from Claude on "best SEO tool for a small agency": hackceleration.com/labs/best/seo-tools-agencies, where Moz came fourth of four. Citations move too: onelittleweb.com fell to 6 answers in pass 2.
• 4. Exact sentences. 25 sentences named Moz in pass 1 (32 in pass 2), copied word for word. One, from Claude: "Moz Pro – best for small businesses, freelancers, and SEO teams who want reliable keyword tracking, link analysis, and site auditing at a lower price than Semrush or Ahrefs."

Why do we not score sentiment?

Because a score needs a judge, and we chose not to use an AI as one. We copy the exact sentences; the client reads the tone and does not have to trust a number.

AgencyAnalytics lists visibility, position, citations and sentiment. We report the first three as counts and show sentences for the fourth: the Claude sentence above praises Moz and also calls it the cheaper option, which one mood score would hide.

What do these numbers not tell you?

They do not show traffic, conversions or why traffic fell. They show where the brand stands in the assistants' answers on a given date.

• No traffic. A named brand is not a click. Put AI-referred visits and conversions from the client's own analytics next to these four.
• API, not app. The answers can differ from what a person sees in the app.
• Not Google. For AI Overviews and AI Mode, use the Generative AI performance report in Search Console. Google warns that "No third-party tool has access to our internal ranking or AI systems."

How do you lay it out for the client?

Start with one sentence, then the four numbers, then the lost questions. The Moz sample report runs 3 pages.

1. The sentence. "Across two runs, Moz is named in 14 to 18 of 40 answers; Semrush, in 36 to 39."
2. The brand against 3 competitors, per assistant.
3. The lost questions, with the page cited in each, for example "best rank tracking software": SE Ranking first on 4 of 4 assistants in pass 1, 3 of 4 in pass 2.
4. The sentences, for the brand and its top competitor.
5. The change since last month, read against the gap between runs: "no clear change" when that is the truth.

By hand, the full report is 30 questions, 3 times each, on 4 assistants: 360 answers to ask, read and tally per client. Sillage Data's full report does it white-label, under your logo, for $349, paid after you have it, and nothing to pay if it is not delivered within 3 business days after we have the brand, its website and 3 competitors (Refund Policy). Monthly monitoring, up to 5 brands, is $549 a month. Start with the free sample.

Questions agencies ask

Which AI visibility metrics should an agency report to a client when organic traffic drops?

Show what you can measure: the share of AI answers that name the brand, the place, the pages cited and the exact sentences, next to the client's conversions. Do not promise a recovery.

Is a one-month change in AI visibility real?

Not necessarily, and we have not measured month-to-month drift. In our two Moz passes a few hours apart, with the same questions and models, Moz's count went from 14 to 18 of 40 answers. Keep the questions fixed, run them more than once, repeat monthly, and call a change only when it stands out from the run-to-run gap.

How often should an agency report AI visibility?

Every month, with the same questions, rules and assistants, so a change shows up as a change. One run is a snapshot.

Can these numbers cover Google AI Overviews and AI Mode?

No. These counts come from the assistants' APIs, not from Google. For AI Overviews and AI Mode, use the Generative AI performance report in Search Console; Google says that "No third-party tool has access to our internal ranking or AI systems."

How we tested and sourced this article

Sillage Data sells white-label AI visibility reports to SEO agencies; this test is its own. The Moz example is a real run of our report on October 29, 2026: 10 questions on ChatGPT (gpt-6-luna), Claude (claude-sonnet-5), Gemini (gemini-3.6-flash) and Perplexity (perplexity/sonar), through their official APIs with web search on, US market requested (Gemini's API does not take a country). Pass 1 ran between 16:41 and 17:03 UTC; the tables other than the two-pass table come from it, as does the sample report. Pass 2 repeated the same questions and models at 21:24 UTC. All 80 answers were read by code: brand matching, order (among the four brands tracked) and cited sites, with no opinion scoring. Of the 6 questions a competitor won, "Semrush alternatives" names Semrush, so 5 if you set it aside. Moz is a public brand used as an example, not a client.

Check it yourself. The raw answers are public: pass 1 (JSON), one record per answer, and pass 2 (JSON), with the full text of each answer. Run the test yourself: our tool is public on Apify.

AgencyAnalytics: Organic Traffic Dropped? What to Tell Clients & the AI Visibility Metrics to Report Instead (Search Engine Journal, August 11, 2026, a sponsored article by the AgencyAnalytics CMO, whose survey figures we could not verify independently). Google: Optimizing your website for generative AI features on Google Search. We are not affiliated with Moz, AgencyAnalytics, OpenAI, Anthropic, Google or Perplexity.

Related reading

• How to show a client whether ChatGPT, Claude and Gemini name their brand
• ChatGPT vs Claude vs Gemini: do they recommend the same brands?
• What to say when a client's ChatGPT check disagrees with your report
• Best white-label AI visibility reports for SEO agencies