# How do we build a simple dashboard to track AI recommendations?

Canonical: https://fortitudemedia.ai/insights/build-ai-visibility-dashboard
Author: Ashley Kenerson
Published: 2026-03-31
Last reviewed: 2026-09-24
Reviewed by: Ashley Kenerson
Pillar: Costs, returns and the business case
Summary: How to build a simple AI visibility dashboard in a spreadsheet, with formulas, and turn it into a one-page board report: what to track and what to leave out.
Keywords: ai visibility dashboard, ai search reporting template, track chatgpt mentions in a spreadsheet, ai share of voice kpis, monthly ai search report for the board, looker studio ai visibility
Short answer: Use a shared spreadsheet with four tabs: an answers log, enquiries, a work log and a summary. Each month, ask the same 20 to 50 buyer questions in ChatGPT, Gemini, Copilot, Claude and Perplexity, several times each in a fresh chat, and count a win only when it repeats. Keep eight to twelve numbers on the main view, with enquiries and wins at the top and GA4 visits as supporting evidence. Give it one owner, and turn it into a one-page board report each month.
What changed: Merged the monthly board report guide and added formulas, GA4's AI Assistant channel, Search Console AI reports, named trackers and 20 to 50 questions counted on repeat.
Publisher: Fortitude Media Limited

When quoting, attribute to Fortitude Media and link to the canonical URL above.

---
You have started the work. The website has new pages answering buyer questions, someone is pitching comment to the trade press, and now and then a prospect mentions they found you through ChatGPT. Then the leadership meeting asks whether it is working, and the evidence is scattered across five places: the CRM, Google Analytics, a spreadsheet of AI answers someone keeps when they remember, an agency PDF and the sales team's memory.

Your usual reports will not settle it. A ranking report no longer tells you who AI assistants name: [Ahrefs found in March 2026](https://ahrefs.com/blog/ai-overview-citations-top-10) that only 38% of Google AI Overview citations came from pages ranking in the top ten, down from 76% in July 2025. And the numbers that do matter, such as how often assistants name you and how many enquiries mention them, sit nowhere by default.

A dashboard fixes that. It is not software you need to buy. It is one place where a handful of numbers are recorded the same way every month, so you can see the trend and decide what to do. This guide shows what to track, how to build it in a spreadsheet in an afternoon, with the formulas, and how to turn it into a one-page monthly report the board will read.

## Key takeaways

- A useful AI search dashboard tracks four things: enquiries and wins, how often you are named, accuracy, and the work done.
- Start in a shared spreadsheet, because most founder-led firms never need anything more elaborate for this job.
- Fix a list of 20 to 50 buyer questions, ask each several times a month in a fresh chat, and count a win only when it repeats.
- Keep eight to twelve numbers on the main view, with results at the top, and show each as a count with a three-month average.
- Treat GA4's AI Assistant channel and Search Console's AI impressions as supporting evidence, because most AI-influenced buyers never click a link.
- Turn the dashboard into a one-page board report every month, and report a bad month plainly with a cause, a fix and a date.
- Give the dashboard one named owner and a fixed update day, or it will go stale by the third month.

## What is the dashboard for, and how is it different from the board report?

The dashboard is the working record: the numbers behind the story, including the question-by-question log. The board report is the one page you take to the meeting, picking out what matters this month. Keep them separate but built from the same data.

The dashboard has three jobs. It shows whether you are being named more often by AI assistants. It shows whether that is turning into enquiries and clients. And it records what was done, so you can connect cause and effect when something moves. If a number does not help you decide what to do next month, leave it off.

## What should the dashboard track?

Four groups of numbers, in this order of importance, plus two supporting figures.

| Group | Numbers to record | Why it matters |
|---|---|---|
| Results | Enquiries that mentioned an AI assistant, how many were qualified, their pipeline value, clients won | These justify the budget, though they may be zero for the first few months |
| Recommendations | Tracked questions each assistant names you for, questions where you are named first, the same count for your leading competitor | These move before enquiries do |
| Accuracy | Answers that describe your firm wrongly | One wrong answer repeated to every buyer can cost more than being absent |
| Work done | Pages published, coverage earned, listings corrected, new reviews and case studies | Inputs that explain movement above, so they sit at the bottom |
| Supporting | Visits from AI assistants in GA4, and AI Overview and AI Mode impressions in Search Console | Useful context, never the headline |

Our guide to [why ChatGPT describes a firm wrongly](/insights/ai-conflicting-information-online) explains where accuracy errors come from and how to correct them.

## How to build it, step by step

You can build a working version in an afternoon. The discipline is in recording the same things the same way every month.

### Step 1: Fix the buyer questions you will track

Choose 20 to 50 questions your buyers ask, in their words, taken from sales calls and enquiry emails. Mix shortlist questions ("best employment lawyers for tech firms in Leeds") with problem questions ("how do we handle a grievance against a director?"), and never include your own firm's name. Once the list is set, do not edit it; add new questions in a separate section. [Tracking whether AI tools recommend your business](/insights/track-ai-recommending-your-business) has more on choosing them.

### Step 2: Set up four tabs

- **Answers log.** One row per question, per assistant, per month, with columns for month, question, assistant, run 1, run 2, run 3 (Y or N for named), named on repeat, named first, accurate, and competitors named.
- **Enquiries.** One row per enquiry: date, source, whether AI was mentioned, qualified or not, pipeline value and outcome. Link to your CRM if it holds this.
- **Work log.** One row per piece of work: what, where and when.
- **Summary.** The main view, calculated from the other three.

### Step 3: Add the formulas

In Google Sheets or Excel, give the month, assistant and named-on-repeat columns the names Month, Assistant and Repeat (in Google Sheets, Data, then Named ranges; in Excel, Formulas, then Define Name) so formulas stay readable. Three formulas do most of the work:

- **Named on repeat**, in each row: `=IF(COUNTIF(D2:F2,"Y")>=2,"Y","N")`. A question counts only if you were named in at least two of three runs, because answers vary from run to run.
- **Share of answers** for one assistant in one month: `=COUNTIFS(Month,"2026-09",Assistant,"ChatGPT",Repeat,"Y")`, shown beside the number of questions asked, never as a bare percentage.
- **Three-month average** on the summary tab: `=AVERAGE(B5:D5)` across the last three monthly columns.

That is the whole template. You can copy the column list above into a blank sheet and be recording within the hour.

### Step 4: Record the answers each month

Put every tracked question to ChatGPT, Gemini, Copilot, Claude and Perplexity, and check whether Google shows an AI Overview or AI Mode answer that names you. Use a fresh chat or private session each time. This is how we track client results: for ID-Pal, "The same buyer questions run every month in ChatGPT, Gemini and Claude, in a fresh chat. A win only counts when it repeats, and losses are reported too." By hand this takes a few hours a month; a tracking tool cuts it to minutes.

### Step 5: Capture where enquiries come from

This is the step most firms skip and the one the board cares about most. Add "How did you first hear about us?" with an AI assistant option to your enquiry form, ask the same on the first call and record the answer in the CRM. Our guide to [tracking which leads came from ChatGPT](/insights/attribution-ai-era-track-leads) covers the detail, including AI-sourced enquiries that arrive looking like Direct traffic.

### Step 6: Pull the website figures

Since 13 May 2026, GA4 has had a default [AI Assistant channel](https://support.google.com/analytics/answer/9756891?hl=en) covering ChatGPT, Gemini, Copilot, Grok and DeepSeek, as [Nice Looking Data reported in July 2026](https://www.nicelookingdata.com/blog/ga4-ai-traffic-chatgpt-referrals). Perplexity and Claude still appear in Referral, so add them with a custom channel group or note them separately. Clicks from Google's AI Overviews and AI Mode count as Organic Search. For those, use Search Console's [generative AI performance reports](https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports), live for all sites since 31 August 2026, which show impressions and pages, though not clicks. Keep each to one line.

### Step 7: Build the summary view

Results at the top, recommendations beneath, then accuracy, work done and supporting figures. Show this month, the three-month average and the year to date. Add a short box for commentary: what changed and why. Colour a cell green, amber or red only where there is an agreed target.

## What should the main view look like?

Here is an example for a professional services firm six months in. Every figure is illustrative.

| Measure (example figures, 30 fixed questions) | This month | 3-month average | Year to date |
|---|---|---|---|
| Enquiries mentioning AI | 2 | 1.7 | 9 |
| of which qualified | 2 | 1.3 | 6 |
| Pipeline from those enquiries | £35,000 | n/a | £140,000 |
| Clients won | 0 | n/a | 2 |
| Named by ChatGPT | 9 of 30 | 8 | n/a |
| Named by Gemini | 7 of 30 | 6 | n/a |
| Named by Copilot | 5 of 30 | 5 | n/a |
| Named by Claude | 6 of 30 | 5 | n/a |
| Named by Perplexity | 11 of 30 | 10 | n/a |
| Named in Google AI Overviews | 4 of 30 | 4 | n/a |
| Named first, any assistant | 4 of 30 | 3 | n/a |
| Leading competitor named, any assistant | 20 of 30 | 20 | n/a |
| Answers with a wrong fact about us | 1 (old service line, correction under way) | 1 | n/a |
| Work: pages, press comments, listings, reviews | 2, 1, 3, 4 | n/a | n/a |
| Supporting: visits from AI assistants in GA4 | 46 | 38 | n/a |

Notice what it leaves out: word counts, social impressions and scores from SEO tools.

## Which tool should we use?

Start with Google Sheets or Excel. For most founder-led firms with £1m to £20m+ revenue, a well-kept spreadsheet is all this dashboard needs to be.

| Option | Good for | Watch out for |
|---|---|---|
| Google Sheets or Excel | The dashboard itself; free or already paid for | Needs one owner and a fixed update day |
| Looker Studio | Charts that update from the sheet and GA4 | Worth it only once the sheet is stable |
| AI visibility trackers, such as Otterly, Peec AI, or the AI features in Semrush and SE Ranking | Running the questions for you across assistants and competitors | Check which assistants each covers, how many runs per question, and that you can export the raw answers |
| Search Console and GA4 | Google AI impressions and AI referral visits | Supporting evidence only |

The hard part is collecting honest data every month, not drawing charts.

## How do we turn the dashboard into a board report?

Take one page from it each month, in six short sections: status, results, leading indicators, what we did, what did not go to plan, and next month. Results come first because that is what the board came for; activity comes last because it only explains the rest. Using the example above:

| Section | What it says (example) |
|---|---|
| Status | Amber. Named more often; enquiries steady. |
| Results | 2 enquiries mentioned an AI assistant (last month 1), both qualified; £35,000 of pipeline; 2 clients won from AI-sourced enquiries this year |
| Leading indicators | Perplexity names us for 11 of 30 fixed questions and ChatGPT for 9 (three-month averages 10 and 8); leading competitor named for 20; one answer lists an old service line, correction under way |
| Supporting evidence | 46 visits from AI assistants in GA4 (three-month average 38) |
| What we did | 2 pages published, 1 trade press comment, 3 listings corrected, 4 new reviews |
| What did not go to plan | Manufacturing case study delayed by client approval; now due next month |
| Next month | Publish the case study; pitch two trade titles; no board decision needed |

Helm Club's weekly numbers work the same way: "Qualified calls a week, reported on one page every Monday and checked against Helm's own member records rather than a platform dashboard." ([Helm Club case study](/work/helm-club)) One page, your own records, the same measures every time.

### How to report a bad month

Plainly, early and with a plan, using the same four lines every time (example):

- **What happened.** "We were named for 7 of 30 questions, down from 11."
- **Why.** "Two answers now cite a competitor's new salary survey, which replaced our older guide."
- **What we are doing.** "Updating our guide with this year's figures and pitching the findings to a trade title."
- **When we will know.** "By the December report. If not, we will bring a revised plan."

If a number is down three months running, that is not noise. It is a trend, and the board should hear it from you.

### How to run the monthly review

Twenty minutes on the leadership agenda is enough. Send the page two days before. Spend two minutes on status and results, and ask the sales lead whether the AI-sourced enquiries were good ones. Discuss one issue, not ten. Record who does what by the next report. Every third month, add the full competitor comparison from [benchmarking your AI share of voice](/insights/benchmarking-ai-visibility-competitors) and decide whether to keep, change or grow the budget.

## Mistakes that make a dashboard useless

- **Too many numbers.** If the main view does not fit on one screen, cut it.
- **A changing question set.** Edit the questions and every trend line becomes meaningless.
- **Percentages without counts.** "Named in 30 per cent of answers" hides whether that is 3 of 10 or 15 of 50.
- **Counting single runs.** One good answer is not a win until it repeats.
- **Effort dressed as results.** Pages published is not a result. Put it at the bottom.
- **No owner or fixed day.** Dashboards die from neglect, usually around month three. Name one person and put the update in the diary.

## What to do this month

1. Agree 20 to 50 buyer questions with your sales team and lock the list.
2. Add an AI assistant option to the "How did you first hear about us?" question on your form and in the CRM.
3. Set up the four tabs and formulas, and record this month's answers as your baseline.
4. Name one owner and a fixed monthly update day.
5. Take the first one-page report to your next leadership meeting, even if most results are still zero.

## Frequently asked questions

### Is there a free template for tracking AI visibility?

This guide gives you one: the four tabs, the answers log columns in Step 2 and the formulas in Step 3. Build it in a blank Google Sheet or Excel file. Keep it in a file your firm owns, not in an agency's tool, so the history stays with you if you change provider.

### Can we build an AI visibility dashboard in Looker Studio?

Yes. Looker Studio is Google's free reporting tool and can read a Google Sheet and GA4 directly, so the sheet stays the source and Looker Studio draws the charts. Do it once the spreadsheet has run for two or three months, because charts built on an unstable sheet break.

### Which KPIs matter most for AI search?

Enquiries and clients that came through AI answers, then the share of your fixed questions where you are named on repeat, compared with your leading competitor. Accuracy comes next. Visits from AI assistants and Search Console AI impressions are supporting figures.

### What is a good AI share of voice for a B2B firm?

There is no published benchmark we trust for B2B services, and it depends on how crowded your market is. Compare yourself with your leading competitor on the same questions, and judge your own three-month trend. A rising share against a steady competitor is a good sign.

### How do we track Google AI Overviews for our key searches?

Search each tracked question in Google and record whether an AI Overview appears and whether it names or cites you. Search Console's generative AI reports then show which of your pages appeared in AI Overviews and AI Mode, and how often. Some trackers also check AI Overviews for you.

### How much time does the dashboard take each month?

By hand, a few hours: most of it running the questions and recording the answers, then half an hour on enquiries and commentary. A tracking tool cuts the answers work to minutes. Enquiry data takes little time if your team records the source when each enquiry arrives.

### Should we track Google rankings on the same dashboard?

One line is fine if your team already watches rankings, but keep it below the AI measures. Rankings and AI answers can move independently, so a steady ranking can hide falling visibility in AI answers, and the reverse.

The first number on any dashboard is your starting point. If you want a baseline before you build the spreadsheet, [book a free Growth Review](/growth-review): 30 minutes with David Adams, and three fixes in writing afterwards.

## Sources

- [Ahrefs, AI Overview citations from top-ten pages (March 2026)](https://ahrefs.com/blog/ai-overview-citations-top-10)
- [Google Analytics Help, Default channel group (September 2026)](https://support.google.com/analytics/answer/9756891?hl=en)
- [Google Search Central, Introducing Search generative AI performance reports in Search Console (June 2026)](https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports)
- [Nice Looking Data, AI traffic in GA4: how ChatGPT, Perplexity, Gemini and Google AI Mode referrals show up (July 2026)](https://www.nicelookingdata.com/blog/ga4-ai-traffic-chatgpt-referrals)