AI Visibility Dashboard: The Weekly Report Your CMO Reads

An ai visibility dashboard turns three metrics - citation share, share of model and AI share of voice - into one weekly page your CMO reads in ninety seconds. You build the ai visibility dashboard by running a fixed panel of 40-60 buyer prompts across ChatGPT, Perplexity, Gemini and Grok on the same day each week, computing the scores with exact arithmetic, then presenting three headline tiles, a competitor bar, an engine breakdown and a three-sentence narrative explaining what moved. This guide walks the full build, template included.

Why your CMO ignores AI visibility numbers today

I've watched sharp teams do exactly this. Someone runs twenty prompts through ChatGPT, screenshots the answers, pastes them into a doc titled "AI Visibility Check" and drops it in the leadership channel. Silence follows, and it has nothing to do with leadership caring too little. The numbers arrive naked - no frame, no trend line, no rival to beat. A screenshot of one answer is an anecdote, and executives act on scoreboards.

The fix is editorial as much as analytical. One page. Three metrics. A delta from last week. A single sentence explaining what moved. That format is the whole difference between a report that earns a standing slot in the Wednesday deck and one that dies quietly in a shared drive.

The stakes deserve a concrete picture. Imagine a brand that ranks page one on Google for its core category yet shows up in two of forty assistant answers to the same buying questions. Every organic report the team produces will glide straight past that gap, and it is precisely the gap a CMO wants surfaced before a board member asks about it.

What goes into an ai visibility dashboard?

Three headline metrics, and I mean only three. Everything else - engine breakdowns, prompt-level detail, competitor movement - sits below the fold as supporting evidence for the tiles above it. The discipline matters more than any individual number, because a dashboard with nine tiles is really a dashboard with zero. IBM's framing of AI observability makes the right point: this is a measurement discipline with defined instruments and repeatable methods, and it deserves the same rigor you'd give any other observability practice.

Citation share: the linked-source scoreboard

When an assistant answers with sources, whose pages get linked? Citation share is your cited answers as a share of all answers that include citations. It is the closest thing AI search has to a ranking, because a citation means the retrieval layer trusted your page enough to build an answer on it.

Share of model: what the model believes without retrieval

Strip away the browsing and ask the model directly for a recommendation. Share of model measures how often you surface unprompted, straight from the model's own weights. It moves slowly, which makes it your long-term brand health line and the wrong thing to chase week to week.

AI share of voice: mentions across the full answer set

Mentions are simpler and broader: does the assistant say your name anywhere in the answer, linked or otherwise? AI share of voice is your mentions divided by all brand mentions in the panel. The split earns its place because mentions and citations diverge constantly - a model can know you well and still cite your competitor's comparison page every single time.

How many prompts do you need before the numbers mean anything?

Small samples lie cheerfully. Run ten prompts and your citation share can swing ten points week over week on pure noise, and nothing kills executive trust faster than a metric that whipsaws without a reason. The third time your CMO watches a number jump and fall back with no explanation, the report is finished.

That range is a practical floor, and the reasoning behind it is worth understanding - we walked the statistics in how many prompts you need for trustworthy numbers. Starting from zero, an AI visibility audit is the fastest route to a first panel, because the audit surfaces the buyer questions where you're already invisible.

The fixed prompt panel: your survey instrument

Treat the panel the way a pollster treats a survey. Same questions, same phrasing, week after week. The moment you swap prompts mid-stream, your trend line turns into fiction. When the panel genuinely needs to evolve - a new product line, a new competitor - version it explicitly and note the change on the dashboard itself.

Per-engine sampling: why blended averages mislead

ChatGPT and Grok routinely cite different sources for identical questions, and Perplexity has a retrieval personality all its own. A blended average can sit perfectly flat while one engine collapses and another surges underneath it, hiding the exact movement you needed to see. Sample at least three engines and report each as its own tile.

Step 1-3: collect answers, extract entities, compute the scores

With the panel in place, the pipeline itself is straightforward. Collection, extraction, computation - in that order, every week, with no improvisation in between.

Collection: same prompts, same day, every engine

Run the full panel on the same day each week so your deltas compare like with like. Capture the complete answer text plus every cited URL, because you'll want to re-classify historical answers when your rules evolve.

Extraction: mentions, citations and the gray areas

From each answer, pull brand mentions and citation domains for you and every tracked competitor. Most of this automates cleanly, though the gray areas deserve a human eye. Does a mention inside a comparison table count the same as one in the recommendation sentence? Does a citation to your docs subdomain count for the parent brand? Decide once, write the rule down, and ratify the edge cases weekly instead of re-litigating them.

Computation: formulas your finance team could audit

The formulas stay deliberately simple. Citation share is your cited answers divided by total answers with citations. AI share of voice is your mentions divided by all brand mentions across the panel. What matters is that the arithmetic is exact - AstroFabric's AI visibility agent runs these calculations in a code sandbox, so the percentage your CMO sees is computed rather than skimmed. When someone asks how a number was derived, you can show the division.

Step 4-5: the template layout your CMO actually reads

Now the editorial work: turning correct numbers into a page someone reads voluntarily. Top to bottom, the layout runs three headline tiles with week-over-week deltas, a competitor bar, an engine breakdown strip, a narrative block, and a cross-channel context row. Here is the whole template as a build sheet.

The three-tile header: metric, delta, target

Each tile carries three things: the current value, the change since last week, and the target. The target is what converts a dashboard from observation into accountability, because "citation share 18%, up 2, target 25 by Q3" is a story with a destination.

The competitor bar: who owns the category's answers

A simple horizontal bar of AI share of voice across every tracked brand. This is usually the element that gets screenshotted into other decks, because nothing motivates investment like watching a rival's bar grow three weeks running.

The narrative block: three sentences that explain the movement

Here is where the report earns next week's readership. One vivid, specific sentence beats a page of caveats: "Citation share up 4 points because our pricing comparison page started appearing in Perplexity answers." That sentence tells leadership what worked, implies what to do more of, and takes five seconds to read. TechTarget's coverage of buyers shifting research into AI assistants is exactly why this block deserves a standing slot in the weekly leadership deck - the behavior change is already underway, and the narrative ties your numbers to it. Anchor the page with a cross-channel row as well, so AI visibility sits beside organic and social share of voice on one scoreboard rather than floating on its own.

Step 6: automate the weekly cadence and delivery

A dashboard nobody opens is a screenshot graveyard. Push the report to where leadership already lives: AstroFabric delivers it by email, Slack or Telegram on schedule, with the console holding the full drill-down for anyone who wants to trace a number back to the answers behind it.

The rhythm that works in practice:

1. Automated collection runs Monday across all engines.
2. A human reviews extraction edge cases Tuesday.
3. The report lands Wednesday morning, before the leadership sync.

Keep the writes approval-gated. The agent drafts the narrative block, a human signs off, and only then does it ship. That single checkpoint keeps the report trustworthy on the week the numbers look bad - which is precisely the week it matters most. Credit-based metering gives the weekly panel run a predictable cost, the kind you can quote to finance in one line and forget about.

Reading the dashboard: what the movements actually tell you

The metrics are the instrument. Interpretation is the skill, and it develops quickly once you know the two patterns that show up most.

Mentions up, citations flat: a retrieval problem

When AI share of voice climbs while citation share holds still, the model knows you but retrieval keeps preferring a rival's page as the source. That points to a content-format problem rather than an awareness one: your pages likely lack the structured, extractable answers retrieval layers reward. If you want to trace one week's number by hand before acting, the worked example in our citation share guide walks a full calculation line by line.

One engine drops: check the pipes before the strategy

A single engine falling while the others hold steady almost always traces back to that engine's crawl or index layer. Run the technical check before anyone rewrites content or convenes a strategy session. Nine times out of ten it's plumbing.

Build yours this week

You now hold the full recipe: a fixed prompt panel, three exactly-computed metrics, a one-page template and a delivery rhythm that puts the report in front of leadership every Wednesday. AstroFabric's AI visibility agent handles the collection, sandbox computation and scheduled delivery, with approval gates keeping a human on the narrative. Start your first panel run and put a real number in front of your CMO next week.

Frequently asked questions

Sources

• IBM - AI Observability
• TechTarget - AI Search Behavior