Cited vs Visible: How to Report ChatGPT Citation Share to Your CMO in 2026

Key takeaways

• Being cited (your URL appears as a source) and being visible (your brand is named in the answer) are different events. Report both, and label which is which.
• Define your denominator before you show a number. Promptwatch Data calculates share only over responses with at least one citation, and your report should state its own rule.
• ChatGPT answers are crowded and unstable. Almost half of cited responses name 10 or more domains, and single runs mislead, so sample each prompt several times.
• Annotate charts with known platform changes. Reddit's ChatGPT citation share dropped sharply in August 2026, and a chart without notes would have looked like a content failure.
• Show your CMO a baseline, a trend, and two or three named competitors. Call the figure directional, not exact.

Why the CMO screenshot problem keeps happening

Here is the usual scene. Your CMO forwards a screenshot of ChatGPT recommending three competitors for a question your buyers ask. You open Search Console and GA4. Impressions are flat, organic sessions are flat. Neither tool says whether an assistant named you last Tuesday.

That gap is why you need a reporting method for AI answers. The numbers exist, but they come from different places, count different things, and use the word "visibility" loosely. If you hand a CMO one blended percentage, the first hard question ("percent of what?") will sink it.

This guide walks through the definitions, the sampling method, the pitfalls, and a one-page report layout. I'll focus on ChatGPT because it is the surface CMOs ask about first, but the same logic applies to Perplexity, Gemini and AI Overviews.

Cited vs visible: the distinction that changes the report

A citation is the assistant linking a specific URL of yours as a source. A mention is the assistant naming your brand in the text of the answer, with or without a link. Visibility is the umbrella term, and it is where reports go wrong.

Assistants name brands more often than they link them. A citation-only view can tell you that you are invisible on a surface where you are actually being recommended. The reverse also happens: your page is cited as a source for a fact, while the answer recommends a competitor by name. A brand can be cited often and still be framed as the cautionary example.

So your report needs at least four separate readings:

The EPR definition of citation share is broader than the strict one above: the percentage of AI answers across a defined prompt set in which a brand is named, quoted or linked. That is a mention-plus-citation blend. Neither definition is wrong. Problems start when two people on your team use different ones.

Digital Applied showed how large the difference can get. On the same data, one brand scored 20% on a mention-based formula, 16.8% position-weighted, and 31.4% citation-based. Pick the formula that matches your objective and write it on the slide.

Pick a denominator and say it out loud

The most common error in AI visibility reporting is an unstated denominator. Share of what? Possible answers:

• All responses in your prompt set
• Only responses that contain at least one citation
• All citations across all responses
• Only citations to a defined competitor set

Promptwatch Data, the first-party research series from Promptwatch, calculates share only over responses with at least one citation, and says so in each report. That is a reasonable default for citation share, because answers without sources would otherwise dilute the number for reasons unrelated to your content. For mention share, use all responses in the prompt set.

Whichever you choose, keep it fixed for the entire reporting period. Changing the denominator halfway through a quarter produces a trend line that is really a definition change.

The ChatGPT denominator is crowded

CMOs often assume an answer cites two or three sources and that you are either one of them or not. The data says otherwise. In Promptwatch's unique domains per ChatGPT response report (Sep 28 to Oct 4, 2026), 47.6% of cited responses cite 10 or more domains, and 66.1% cite at least 8. Only 6% cite three domains or fewer.

The split by prompt type matters for how you frame goals:

Two things follow. First, citation share is divided across many domains per answer, so a share in the low single digits can still be a strong position in a fragmented category. Second, brand-specific prompts are the narrow ones, so do not blend them with organic prompts and call the result "visibility." Report by prompt type.

Promptwatch's own advice from that report is worth lifting into your methodology: track where you appear in the answer and how ChatGPT describes you, check which competitors are cited in the same responses, and measure across many prompts because one prompt tells you little.

Sampling: one screenshot is an anecdote

OpenAI does not publish a citation report for site owners. Microsoft does for Copilot: Bing Webmaster Tools has an AI Performance report with a Citation Share view, which Supermetrics describes in its 2026 measurement guide (Microsoft calls it observational, with no competitor domains and no quality score). For ChatGPT and Perplexity you have two options: a third-party GEO tracker, or a prompt panel you run yourself.

The two methods do not produce equally reliable numbers, so label them. Here is a workable panel protocol:

1. Build 100 to 200 prompts that mirror how buyers ask, split into organic, brand-specific and competitor comparison.
2. Run each prompt three to five times per engine in a clean, logged-out session. EPR cites research finding less than a one-in-100 chance that ChatGPT or Google's AI returns the same brand list twice for the same prompt.
3. Log brands named, order, sources cited, and date.
4. Compute your shares per prompt type, then roll up.
5. Repeat weekly. Digital Applied calls a 100 to 200 prompt panel run weekly the minimum viable system, and notes that cited domain sets drift 40 to 60% month over month in active categories.

Doing this by hand is feasible for a pilot. At scale, tools automate it. Several in our catalog track ChatGPT answers daily, including Otterly.AI and Promptwatch, along with others listed in the GEO software directory at bestgeosoftware.com.

Treat any tracker's numbers as directional. Sampling methods vary, and most vendors do not publish theirs, so a count from one tool is not safely comparable to another's. Pick one and stay with it for the full quarter.

Pitfall: platforms change, and your chart will show it

The best argument for annotating charts is what happened to Reddit in August 2026. According to Promptwatch's Reddit citations data, reddit.com held a steady 3.83% average share of ChatGPT citations from July 18 to August 7. It slipped on August 8, the day ChatGPT changed its query fanout behavior, and fell below 1% on August 14. The August 14 to 17 average was 0.52%, an 86.4% relative drop.

In Google's AI Overviews, over the same period, Reddit eased from 2.37% to 2.10%. Same source, two platforms, different outcomes.

Two lessons for your report:

• If a sudden move shows up in your share, check whether a platform changed. Promptwatch notes that ChatGPT began using the site: operator at scale on August 8, jumping from about 0.4% to about 17% of fanout queries (see the site: operator fanout report).
• Promptwatch also says a data-collection issue cannot be ruled out for the Reddit drop, so check your own collection volume before reading a swing as a change in ChatGPT's behavior.

Put a dated annotation on the chart for every known platform change. Your CMO will forgive a dip with a note. A dip without one reads as a failure of your team.

Citation share is a moving stock, not a snapshot

The third structural fact for your CMO: cited pages decay. Promptwatch's content half-life report tracked every page ChatGPT cited on July 6, 2026 through October 4. After 30 days, ChatGPT still cited 78% of landing pages, 76% of docs and 72% of comparison pages. Product pages, reviews and how-tos sat around 55% after 60 days. News decayed fastest: half of cited news pages were gone after 36 days.

The practical implication is that last month's citation share is partly built on pages that will fall out. Report a rolling trend, not a single point, and segment by content type so the CMO sees which parts of your content are durable. Promptwatch's citation share by content type data shows product pages as the most cited type in ChatGPT (22.4% of citations, Sep 28 to Oct 4, 2026), which helps you decide what to benchmark against.

Domain rank is another trap. Promptwatch's domain rank data shows sites at DR 46 to 75 earning almost half of ChatGPT citations in August 2026, while DR 91 to 100 fell to about 3%. Do not assume that a strong backlink profile will carry you.

Do not collapse engines into one number

A single blended "AI visibility" figure is tempting and wrong. A 2026 per-engine audit cited by Digital Applied and 5WPR found only about 11% of domains cited by ChatGPT overlap with those cited by Perplexity. A tool covering one platform gives a false sense of completeness.

Social sources show how different the engines can be. In Promptwatch's social media citations report, ChatGPT drew 0.46% of its citations from social platforms between Sep 28 and Oct 4, 2026, while Google AI Overviews drew 12.2%. If your strategy leans on community content, one engine will look great and another terrible.

Show per-engine ranges, with ChatGPT as the headline surface if that is what your CMO cares about, and give the composite only as a secondary figure.

What citation share does not capture

Be upfront about the gaps. EPR lists the main ones, and I would put them on a footnote line:

• Accuracy and sentiment. The metric counts presence, not whether the description is correct or favorable.
• Click-through. Most AI answers produce no click. Citation share measures presence in the answer, not what happens next.
• Dark traffic. AI-referred visits are often mislabeled in analytics, so referral counts understate the effect.

To close some of that gap, pair your panel with visitor analytics that separate AI referrals, and with Search Console. Google's generative AI reports (announced June 2026) show impressions for AI features but, per Search Engine Land, no clicks or query data. Bing's report covers Copilot. Neither covers ChatGPT. That is why a tracker or panel is still needed.

Platforms such as Promptwatch also tie visibility to crawler logs and AI-referred traffic, which helps you move from "we were cited" to "it sent visitors." Other vendors cover only the tracking half.

The one-page report your CMO will read

CMOs read market position, not methodology. EPR's advice is to report citation share like market share: a baseline, a trend and a comparison. Here is a layout that fits on one slide:

1. Headline: ChatGPT citation share and mention share for the quarter, each with a range (low and high across repeated runs), not a single point.
2. Competitor row: the same two metrics for two or three named competitors.
3. Trend line: weekly, with dated annotations for platform changes.
4. Prompt type split: organic, brand-specific, competitor comparison.
5. Quality line: percent of answers describing you accurately, plus sentiment direction.
6. Footnote: panel size, runs per prompt, denominator, tool used.
7. Next actions: the three pages or topics you will fix, and what you expect to move.

A small table helps the first conversation:

One more habit: pre-empt the "is that good?" question. Because ChatGPT spreads citations over many domains per answer, benchmark against your named competitors, not against 100%.

Tooling options for the report

You can build this in a spreadsheet, but most teams end up with a tracker. The catalog has dozens, and I would filter on four questions: does it track the real ChatGPT interface rather than only an API, does it show prompt-level responses so you can audit it, does it let you annotate and export, and does it publish its sampling method.

If you want a wider list of tracker options, the AI rank tracking directory at ai-rank-tools.com is a good place to compare.

If you need help with the program behind the numbers, that is what an SEO and GEO agency is for. 1001 SEO Media builds measurement and content work around AI citations, and uses Promptwatch for tracking.

A short checklist before you present

• Is the denominator written on the slide?
• Did you run each prompt several times?
• Are cited and mentioned reported separately?
• Are platform change dates marked on the chart?
• Are ranges shown instead of one decimal-point figure?
• Do you name the tool and keep it fixed all quarter?
• Did you say what the metric does not tell you?

If you can answer yes to all seven, you have a report that survives the first hard question. The number will move next month because the sources drift, and that is exactly why the trend and the footnote matter more than any single percentage.