Reporting AI Visibility to Your CEO: A One-Page Structure

Your CEO wants to know if AI platforms recommend the company and whether that is improving. The honest answer cannot come from a rank, because no rank exists to measure.

As of October 2026, Google's Search Console generative AI report shows impressions only. It has no clicks, no click-through rate and no position. Google also says AI Overviews and AI Mode "may use different models and techniques", so they are two surfaces, not one. A tool that reports "position 3 in Google AI" has nothing to check that number against.

Blended scores have a second problem. Retrieval is per query and discretionary. ChatGPT decides for itself whether to search. Claude.ai search is a user toggle. So any single AI visibility score averages answers given with live web search and answers given from memory alone. When the number moves, you cannot tell your CEO why.

What this means: report facts that are true or false and fixable. A CEO can act on "Perplexity states last year's price". Nobody can act on "AI score down 4 points".

Open with two sentences: what changed since the last report, and the one decision you need from leadership. Then four blocks.

Keep each block to three to five lines. Use green, amber and red for access, a short table for accuracy, and a yes or no grid for presence. If you need a fuller version for your own team, the monthly AI visibility report template expands each block. For what belongs in a longer report and what to cut, see what to include and exclude in an AI SEO report.

1. Check robots.txt per bot. Confirm OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot and bingbot are allowed. Treat GPTBot separately: OpenAI says it governs training, and "each setting is independent of the others". Blocking GPTBot does not remove you from ChatGPT search. Blocking OAI-SearchBot does.
2. Check the firewall and CDN. OpenAI's help center says eligibility for ChatGPT search needs both: allow OAI-SearchBot and confirm your host or CDN allows OpenAI's published searchbot IP addresses. A robots.txt that says allow can sit behind a firewall that returns 403. The AI Crawler Checker requests a page as each AI crawler and shows which ones the firewall turns away.
3. Check rendering. Fetch your pricing and product pages as raw HTML with JavaScript off. Vercel found that ChatGPT and Claude crawlers fetch JavaScript files but don't execute them. If your price only appears after JavaScript runs, those crawlers can't read it.
4. Run a fixed set of buyer questions. Use the same 10 to 20 questions every month, on each engine, logged out with memory off. For each answer, record three things: were you named, were citations shown, and which facts were stated about you.
5. Score the facts. Compare every stated fact against your own source of truth: pricing, products, founding year, leadership, location. Log each wrong one with a severity. Errors about people need their own pass; see checking what AI says about your founders.
6. Pull first-party data. Add impressions from the Search Console generative AI report and citations from Bing's AI Performance report. Label both with what they measure and what they don't.
7. Write the top line last. One change, one decision.

Run through this before the report goes to your CEO.

• Every search crawler (OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot, bingbot) shows a pass or a named problem.
• Any firewall or CDN setting change since last month is noted, including vendor default changes you didn't make.
• Pricing, product names and key numbers are in the raw HTML, not only after JavaScript.
• No page you want quoted carries noindex, nosnippet, max-snippet:0 or data-nosnippet by accident.
• Each wrong fact lists the engine, the date, what it said, the correct fact, and the fix status.
• Each presence row says whether citations appeared, so the reader knows if the engine searched.
• The same questions were used as last month, or the change is stated.
• No rank, no position and no single cross-platform score appears anywhere.
• Share metrics carry a plain label. Read citation rate vs citation share before you put either in front of an executive.
• The "outside our control" block is filled in, not left blank.

A good month

• All search crawlers return a normal page, and robots.txt matches what the firewall actually does.
• Zero critical accuracy errors open. Older errors show as fixed, with the date fixed.
• Presence is reported per engine, with retrieval state shown, for example: named with citations on Perplexity, named without citations on Claude.
• The top line names a specific cause, such as "pricing moved into server-rendered HTML on 3 October".

A bad month, and what to do

• A 403 to a search crawler. This is a silent, complete failure for that engine. Fix it first and say so in the top line.
• A wrong price repeated across engines. Check whether your own site states it two ways. Fix the contradiction on your pages before anything else.
• Presence dropped on one engine only. Check that engine's access before concluding anything about content.
• The report leans on a rank or a blended score. Remove it. If the CEO has been seeing one, explain in one line why it is gone.

A month with no change in presence is not a failure if access is clean and errors are falling. Those are the parts you control.

This block protects your credibility. It tells the CEO which movements are not your team's doing.

• Model memory. When an engine answers without searching, it uses what the model learned before its knowledge cutoff. Nothing you publish this quarter changes that answer.
• Whether the engine searches at all. ChatGPT decides per query. Claude.ai search is a user setting.
• The queries the engine actually runs. OpenAI's help center says ChatGPT search typically rewrites your question into one or more targeted queries. Your test questions are a proxy for inputs you cannot see.
• Which sources get chosen. A July 2026 critical survey of 45 GEO studies found no technique with a stable, cross-platform causal effect on discoverability.
• Third-party pages. When engines cite other sites, being named on those sites is associated with appearing in the answer. You can influence that, but you cannot control it.

Say plainly that no one can promise citations. Then point back to what the team did fix.

Use three rhythms.

• After any deploy or CDN change: recheck access and rendering. A firewall rule or a CDN default can block AI crawlers without anyone editing your site.
• Monthly: send the one-page report with the same question set.
• Quarterly: show the trend in fixed errors and per-engine presence, and refresh the question set if your products changed.

Answers vary between runs, even with nothing changed on your side. Read what changes between AI visibility checks before you call a one-month swing a trend. If you need a baseline before your first report, the free checker on this site gives a snapshot of up to three buyer questions. It is a starting point, not continuous tracking.