What Your AI-Search Report's Numbers Can Justify

Your monthly AI-search report covers a channel with no line in your sales records. AI visibility and share of voice can each justify one decision; average rank, read alone, justifies none; and none of them turns into revenue without your inquiry records.

On this page
  1. What Each Line on the Report Counts
  2. Why Average Rank Can Rise While You Lose Answers
  3. How Far the Numbers Move When Nothing Has Changed
  4. What Tells You the Channel Sends Buyers
  5. What Qvery’s Agents Put on Your Monthly Page
  6. Your Last Three Reports Are the Starting Point
  7. Frequently asked questions

Say the monthly report on your desk, from an agency, a tool or your marketing lead, on how often AI assistants name you, has three lines on its first page. Visibility is up from 29% to 34%. Share of voice is up two points. Average rank went from 3.2 to 2.1.

The cover note adds one sentence: share of voice is the leading indicator of market share, so next quarter's budget should grow.

The report is about a sales channel that has no line in your customer relationship management (CRM) system. A buyer asks ChatGPT, or Google in its chat-style search answers, which companies handle their problem, and the answer names a few before the buyer contacts anyone.

Whoever sent the report wants a signature on the strength of three lines and one sentence. Your own question is narrower: is that channel sending you buyers, and does the work on it keep its budget?

37.09%of companies whose ChatGPT average rank improved by more than a quarter of a position were named in fewer answers or sat no higher in a shorter list
8.00 pointshow far two askings of the same question list put one company's visibility, one time in ten (ChatGPT; companies named in at least a tenth of the answers)
4 vs 5companies named in the typical ChatGPT and Google AI Mode answer that named any company
Fewer than 50B2B cases behind the rule tying share of voice to market share, all of it measured on ad spend
The GEO Agent's daily records: ChatGPT and Google AI Mode answers, August to September 2026; Les Binet and Peter Field for LinkedIn's B2B Institute

None of the three lines turns into revenue on its own. The rule tying share of voice to market share holds on average, within a narrow scope: it was measured on advertising spend, in fewer than 50 business-to-business (B2B) cases in Les Binet and Peter Field's work for LinkedIn's B2B Institute.

The step from an answer to a buyer is counted in your own inquiry records, which the report does not contain.

The report has a price of its own. If a marketing analyst builds it by hand from the monitoring tool's exports, about 13% of the week, that is about $16,614 a year at the Bureau of Labor Statistics’ mean pay for marketing specialists with benefits.

What Each Line on the Report Counts

A buyer asks ChatGPT which firms handle their kind of problem for a company their size. The answer names four companies, and yours is second. Ask the same question ten times over a month and the answer moves: some days you are in it, some days you are not.

Visibility is the share of the answers that name you at all. Named in four of the ten: 40%. It does not care where you sit or how many others are named.

Average rank is your average position in the answers that name you, and only those. Second all four times: 2.0. The six answers that left you out do not enter it.

Share of voice is a name borrowed from advertising, where it means a brand's share of the media spending in its category.

In an AI report it stands for how much of the answers' attention you get beside rivals: each answer's weight is split among the companies it names, more of it to the ones named first, and an answer that skips you gives you none.

With a simple position weighting of 4, 3, 2 and 1 points, second place in a four-company answer carries 30% of that answer's weight. Over the ten answers, four of them with you second: 4 × 30% ÷ 10 = 12%.

The three numbers side by side
NumberWhat it counts, and out of whatThe question it answers
VisibilityAnswers that name you, out of every answer to the listAre we in the answers at all?
Share of voiceYour position-weighted share, out of every answer (zero where skipped)How much of the answers' attention do we get?
Average rankYour position, out of only the answers that name youWhen we are named, how high?

In the answers Qvery's GEO Agent collected every day in August and September 2026, the typical answer that named any company named 4 on ChatGPT and 5 on Google's, so the same second place was worth 30.00% on ChatGPT and 26.67% on Google. Blending them adds numbers measured on different rulers.

Piotr BabelPartner @ Qvery

Share of voice depends on how many companies each answer names, and that differs by engine: the typical answer named four on ChatGPT and five on Google AI Mode in the GEO Agent's daily records. So we compute every number per engine, on one unchanged question list, and split each month into odd and even days. When the two halves disagree by more than the move you are reporting, the move stays off the first page.

Why Average Rank Can Rise While You Lose Answers

The third line of the report on your desk is the easiest one to misread. Take ten answers a month.

In the first month you are named in seven of them, at positions 1, 2, 1, 2, 4, 5 and 6: visibility 70%, average rank 3.0. In the second month you are named only in the four answers where you were first or second: visibility 40%, average rank 1.5.

Two panels of ten answers each. Month one names the company in seven answers at positions one to six; month two keeps only the four answers where it was first or second.
Look at which rows turned dashed: the low positions, which is why the average improved.Illustration by Qvery, drawn for this article.

Visibility and share of voice both registered the loss that average rank hid.

In the GEO Agent's daily records from August to September 2026, more than a third of the companies whose average rank improved by more than a quarter of a position got there with fewer mentions or no better placement: on ChatGPT 37.09% of them, and on Google 36.79%, were named in fewer answers (by more than two points), or sat no higher in a list that had simply got shorter.

How Far the Numbers Move When Nothing Has Changed

Ask an AI engine the same question twice and you can get two different answers.

With each question asked once and the whole list asked again, the two askings put the visibility of a company named in at least a tenth of its market's answers 8.00 points or more apart one time in ten on ChatGPT.

With each question asked every day for a month and the answers split into odd and even days, the two halves landed 2.02 points or more apart one time in ten.

One asking is not a measurement.

What Tells You the Channel Sends Buyers

Drop the borrowed rule and the channel still has an arithmetic. A buyer asks, an answer names a few companies, and some of those buyers contact one of them.

Visibility and share of voice measure the middle step only. How many buyers asked is not in the report, and neither is how many of them reached you.

Vlad ShvetsPartner @ Qvery

The cover note's rule, that share of voice leads market share, rests on fewer than 50 B2B cases measured on advertising spend, so it cannot carry a budget increase on its own. Sign the bigger budget only when the inquiries whose source answer names an assistant rise beside visibility. If your inquiry form doesn't ask that yet, add the answer this month and judge the channel next quarter.

Follow the inquiries whose source answer names an assistant through to qualified deals and signed work: that is the arithmetic the cover note skipped.

The job belongs to whoever owns the pipeline, the person who answers for new business, because the decisions are theirs and so is the inquiry form.

A number that cannot name the decision it supports does not belong on the first page of the report.

What Qvery’s Agents Put on Your Monthly Page

The agents sort inquiry answers such as "my colleague asked ChatGPT" into the count and read the answers behind each real move. The arithmetic needs no agent; a script computes each number twice, on odd and even days.

Four steps: export the answers, inquiry records and CRM source field, compute each number twice with a script, the agents quote the answers behind each real move, the pipeline owner signs off.
The split is the design: arithmetic to a script, reading to the agents, the decision to a person.Illustration by Qvery, drawn for this article.

Most of the volume is in the reading. An example list of 50 buyer questions, asked daily on both engines, is 3,000 answers a month, far more than anyone can read for a one-page report.

What a team does by hand today is the report itself. The calculator prices it for a company of about 250 people: one marketing analyst turning the monitoring tool's exports into the monthly slides and commentary and setting them beside the inquiry records, about 13% of the week, about $16,614 a year.

With the agents taking about 65% of it, at about $250 a month to run, it costs about $8,815, saving about $7,799 a year and 176 hours; each $10,000 of build cost is repaid in about 15.4 months. The share the agents take is set when we build them; it and the running cost are our assumptions, to replace with your own.

Monthly AI-Search Report ROI Calculator

What Turning The Monitoring Tool's Exports Into The Monthly Report: The Slides, The Commentary And The Comparison With Your Inquiry Records Costs You, And What An Agent Saves

The marketing analyst who builds the monthly report. One is our assumption for a company of about 250 people.
Pay plus benefits. Market research analysts and marketing specialists earned a mean $89,490 in May 2025, the Bureau of Labor Statistics reports; with benefits at 30% of private employer costs, about $127,800.
13% is about 23 hours a month: exporting the monitoring tool's numbers, building the slides, setting them beside the inquiry records and writing the commentary, by hand. An assumption.
Building the page from the export and drafting the commentary, about 65%, an assumption. The person responsible for the monthly report approves every page; the budget decision stays with people.

Saved Per Year

$7,799

176 hours handed back, after $3,000 a year to run it

Hours Handed Back
176 hours a year
Agent Running Cost (Estimate)
$250 a month
Payback Per $10,000 Of Build
15.4 months

On these numbers, the agent saves $7,799 a year after its own running cost, and every $10,000 of the build is earned back in 15.4 months.

Example numbers. Type over them with yours.

The pipeline owner checks the draft before the budget conversation. The agents change only the draft: they publish nothing, edit no web page and send no message.

For the monthly report, we learn how your pipeline owner checks the export and inquiry records, then fit custom agents into the systems holding them. Their access stops at the answer export and the CRM's source field, and every move they flag is logged with the answers it rests on.

Your Last Three Reports Are the Starting Point

We start from your last three monthly reports and the inquiry form's source answers, so the first page the agents draft shows which of the old lines held up. Two weeks after they can read the export and the CRM, your pipeline owner has that page to check before the next budget conversation.

Frequently asked questions

What is the difference between AI visibility and share of voice?

Visibility is the share of AI answers to your buyer questions that name you at all. Share of voice is your share of the attention in those answers, each mention weighted by position, with answers that skip you counted as zero. It equals visibility times how prominent you are when named.

Does a higher share of voice in ChatGPT mean more sales?

The research behind the share-of-voice rule does not establish that link: it measured advertising spend, in fewer than 50 B2B cases, and its authors call the results tentative. Ask new buyers how they heard of you, then follow inquiries naming an AI assistant through to qualified deals and signed work.

Why did our average rank improve while we were mentioned less?

Average rank counts only the answers that name you, so dropping out of the ones where you sat low improves it. A shorter list can improve it too, when the companies dropped sat above you. From August to September 2026, more than a third of the improved average ranks in the GEO Agent's daily records came with fewer mentions or a shorter list.

What is a good AI visibility score?

No score transfers from one company to another: visibility depends on the questions, their wording and the engine. Compare your number with itself, per engine, on an unchanged list, against how far it moves when nothing has changed. Questions that contain your company's name produce a flattering number that says nothing about buyers.

How much does a number have to move before it means something?

It depends on how often each question was asked. In the GEO Agent's daily records, one asking per question put the same company's visibility 8.00 points or more apart one time in ten on ChatGPT; daily askings for a month, 2.02. To see how far your own numbers drift, ask your list twice in a week.

Can I compare our share of voice on ChatGPT with Google AI Mode?

Compare each engine with its own history first: in the records analyzed here, the typical Google AI Mode answer named five companies and the typical ChatGPT answer four. The same second place was worth 26.67% and 30.00%, so a gap between engines partly measures how their answers are built.

How do we know whether AI answers are sending us buyers?

Add "An AI assistant such as ChatGPT or Google AI Mode" to your inquiry form’s how-did-you-hear question and ask it on the first call too. Count those answers beside visibility each month, treating them as the buyer’s account, and follow the inquiries to qualified deals.

Who in the company should own AI-search measurement?

Whoever owns the pipeline: the person who answers for new business and controls the inquiry form. Collecting the answers, computing the numbers and drafting the page can go to a tool and agents. The question list and the decision stay with that person, who alone can set the numbers beside the inquiries.

Do these numbers show how many buyers saw us?

These numbers describe the answers to your question list, not how many buyers saw you. They show what a buyer would read if they asked; none measures how many people asked or read the answers.

What should happen when our tracked question list changes?

Count the changed questions afresh rather than reporting their difference as progress. Otherwise, the report cannot distinguish a change in the answers from a change in what you asked.

Which number belongs on the first page of our report?

The one tied to a decision: visibility, per engine, set beside the count of inquiries that name an assistant. Average rank belongs only beside visibility, because read alone it can improve while you lose answers.

Keep Reading

35 Domain Authority and ChatGPT Statistics for Software (2026)

35 statistics on whether domain authority predicts which software brands ChatGPT names: the cited best-of lists tracked it far more closely.

45 Law Firm AI Search Statistics (2026)

45 law firm AI search statistics: ChatGPT opened 83.72% of lawyer answers with a rated card, and 33% of firms answered an emailed inquiry.

50 Personal Injury AI Search Statistics for Law Firms

50 personal injury AI search statistics: accident questions and lawyer questions barely share a source (1.70% and 7.65%).