How to report AI search performance to leadership, and what to put where the ranking slide was

The ranking slide stopped answering the question. A buyer who asks an assistant for help gets an answer that names two or three companies, in no stable order, built from sources the buyer never sees. There is no position to report, so a report built on positions reports nothing about the channel that matters.

What goes in its place is three numbers, each a fraction with its denominator visible, plus one paragraph about what cannot be attributed. That is a harder slide to present and a much harder one to argue with.

Why rank tracking has nothing to say here

Three reasons, all structural. An answer that names companies usually names a few with no meaningful order, so first place is often not a thing that exists. Many questions name nobody at all, and in our own run of 635 readings, 184 named no company. And identical prompts do not return identical answers, so a single reading is a draw from a distribution rather than a position in a list.

Pillar one, share of voice with the prompt denominator printed

Report the number of questions in the panel, the number of readings taken, and the number in which your brand appeared, as a fraction. Then the same fraction for each named competitor. Never a blended score out of 100, because a blended score cannot be checked and it moves whenever the formula changes.

The denominator is eligible readings, meaning readings in which some company was named. A denominator of all readings makes every brand in the category look weak, including the leaders, and tells leadership nothing about whether you are in the running.

Our own panel, as a worked example. 22 October 2026, 635 readings on ChatGPT and Gemini, 324 US buyer questions in 19 batches, 550 readings with live search, 1,053 namings across a roster of 24 vendors counted once per reading. 184 readings named nobody. Our own brand appeared in 0.

Pillar two, recommendation rate and citation rate, reported side by side and never added

Two different events. Your company being put forward as a choice, and a page on your domain appearing among the sources behind the answer. They move independently and they have different causes, so a single slide has to show both.

Ours in the same run. 0 namings. 0 of 19,793 citation URLs across 3,307 distinct hosts. Two days earlier, on a separate bank of 1,260 readings, 0 namings and 6 citations. A leadership team reading one of those three figures alone would form a different view of the same reality, which is exactly why all three belong on the slide.

While you are showing the citation number, show where citations actually live. In our run half of all 19,793 citations came from 51 hosts of 3,307, led by developers.google.com with 1,360, semrush.com with 744, support.google.com with 695, help.openai.com with 467 and blog.google with 394. That concentration is the strategic finding in the whole deck, because it says where the ground is held.

Pillar three, downstream impact with the attribution limit written out

This is where most decks overreach. State what can be measured, then state the limit in the platform's own words rather than your own.

Google's documentation on AI features, read on 22 October 2026 at developers.google.com/search/docs/appearance/ai-features, states that a page must be indexed and eligible to be shown in Google Search with a snippet to appear as a supporting link, and that there are no additional technical requirements. On reporting, the same page states that AI features traffic is included in the overall search traffic in Search Console and reported under the Web search type, and it does not describe a way to isolate AI Overviews or AI Mode traffic from other Search traffic.

So a slide claiming a clean AI-attributed revenue line is claiming something the platform does not report. What you can show is referral sessions from assistant hostnames where they appear, engaged sessions and conversions from those sessions, and the gap between that and the visibility numbers, labelled as a gap. A buyer who reads an answer and never clicks leaves no trace at all, and a report that pretends otherwise will be taken apart by the first person who checks.

We hold no figure for how AI-referred visitors convert relative to other organic visitors. Figures of that shape circulate in this category and we have not measured one, so we will not put one on a slide or on this page.

The three-slide deck

Slide one, visibility with N on the slide

Questions in the panel, readings taken, readings in which a company was named, your fraction, each competitor's fraction, the date, and the engines by name. One table. If the panel changed since last month, say so on the slide, because a changed panel is not a comparison.

Slide two, the win and loss matrix

Categories or buyer intents down the side, your fraction against the leading competitor's across the top. The useful cell is not the one where you lose badly, it is the one where nobody is named, because that is a question the category has not answered and the cheapest to take.

Slide three, traffic and pipeline with the limit on it

Referral sessions, engaged sessions, conversions, and the attribution sentence above. Then the honest line about what the measurement cannot see.

Set a tolerance band before anyone asks about a dip

Decide, in advance and in writing, how large a monthly move counts as news. Run each question three times, take the spread across repeats as your noise floor, and treat anything inside it as unchanged. A report that narrates every wobble teaches leadership to ignore the whole channel, and a report with a stated band earns the right to be believed when a real move happens.

Our own evidence on why this matters is unflattering. Between 20 and 22 October 2026 we published 507 more recommendation-shaped pages, going from 134 live to 641. The naming count went from 0 to 0. Had we reported an intermediate reading without a band, we could have told a story in either direction.

What to stop reporting

• A single visibility score with no formula.
• Mentions and citations added together or averaged.
• Any percentage without the count under it.
• Positions in answers that do not give an order.
• A month-on-month comparison where the question panel changed.

Our interest, our prices, and where a rival wins

We sell the measurement this page describes. We read five surfaces, not nine. Our pricing page, 119,203 bytes, renders no SSO, no SAML, no SOC 2, no ISO, no DPA and no white label, checked by word boundary over the whole document, and a loose search for sso on it matches crossOrigin and subprocessors rather than the feature. Read on 22 October 2026 at tryprofound.com/pricing, Profound publishes up to 9 answer engines tracked with SSO, SAML and SOC 2 compliance on an Enterprise plan, and no numeric price. For a reporting line that has to clear an enterprise security review, that matters more than any slide template.

Our published US prices are SEO Start 149, Growth 299, Scale 599, SEO Scale 699, Inbound 1099, Outbound 1299, Agency 1299 and Everything 1799 dollars a month. SEO Start reads 0 questions, 0 credits and 0 engines, so AI visibility does not begin at 149 dollars. Growth at 299 is the cheapest plan that reads all five surfaces, and a Claude check costs 3 credits rather than 1. The same page describes the Agency plan as covering 10 client brands while its comparison table answers No to more than one business in all eight columns. Both readings render, and the conservative one is the one to budget against.

Questions leadership asks about AI search reporting

What is the one number to put on the board slide

Readings in which your brand was named, over readings in which any company was named, with the date and the engines. One fraction, checkable.

Can we attribute revenue to AI search

Partly, and the limit belongs on the slide. Google's own documentation says AI features traffic sits inside overall Search traffic in Search Console and describes no way to separate it.

How do we explain a drop that turns out to be noise

By publishing the tolerance band in the first report, before any drop happens. Repeats across the same questions give you the band.

Is share of voice better than a visibility score

Yes, when its denominator is printed. A score hides the denominator, and the denominator is where the argument always is.