Cloudflare announced its AEO Visibility Dashboard on 6 August 2026, a product that scores how often a website is cited and mentioned inside AI assistant answers. PPC Land reported the launch the same day, and the single most important detail in it is the scope: the dashboard covers Anthropic's Claude and OpenAI's GPT models only, and does not include Gemini, Perplexity, Copilot or Grok.
That scope decides what the score can and cannot tell a brand. A site can read as strong inside the dashboard and still be absent from the assistants it does not sample, and the product has no way to warn anyone about the gap. Everything else in the announcement sits underneath that limit.
What Cloudflare put in the AEO Visibility Dashboard
Per Cloudflare's 6 August 2026 announcement, as reported by PPC Land, the dashboard reports six things. Five of them describe how a site appears inside answers. The sixth describes traffic between the site and the AI systems themselves.
- Citation Rate: the share of answers that cite the site as a source.
- Mention Rate: how often the brand is named, whether or not the site is cited.
- Prominence: how much of the answer text is attributed to the source, and where in the response it appears.
- Share of Voice: the site's citation share measured against competitors in the same category.
- Industry Fit: whether an assistant places the site alongside genuine competitors.
- AI Operator Activity: crawl and referral traffic broken down by operator, with error patterns.
Stephanie Cohen, Cloudflare's Chief Strategy Officer, framed the launch by saying "Being discoverable used to mean ranking on a page. That's not enough anymore." The company's argument is that recommendation now happens inside an answer rather than on a results page.
The five answer metrics, and what each one is actually counting
The table below restates the five answer level metrics as Cloudflare defined them on 6 August 2026, with a note on what each one changes when it moves. The right hand column is analysis and is not a Cloudflare claim.
| Metric, as defined by Cloudflare | What a change in it would indicate |
|---|---|
| Citation Rate: share of answers citing the site as a source | Movement in how often the site is used as a linked source, the closest thing here to a traffic proxy. |
| Mention Rate: how often the brand is named regardless of citation | Brand recall inside the model, which can rise with no clicks attached to it. |
| Prominence: how much answer text is attributed, and where | Whether the site anchors the answer or is a closing footnote. |
| Share of Voice: citation share against competitors in the category | Relative position, which can fall while Citation Rate rises if rivals move faster. |
| Industry Fit: whether the assistant places the site with genuine competitors | Whether the model has categorised the brand correctly at all. |
Mention Rate and Citation Rate are worth separating in practice. A brand can be named constantly and cited rarely, which produces awareness with no referral path, and the two numbers moving in opposite directions is a real result rather than an error.
Why the two assistant scope is the caveat that matters most
PPC Land's 6 August 2026 report states the coverage plainly: Claude and GPT models, with Gemini, Perplexity, Copilot and Grok outside the product. This section reasons about the consequence, which the source does not spell out.
A visibility score is only meaningful against the population it samples. If a brand's buyers ask Gemini, because it sits inside the Google apps they already use, or Perplexity, because they want sourced answers, then a healthy Cloudflare score is measuring a different audience from the one making decisions. The failure mode is quiet: nothing in the dashboard reads as missing data, because the assistants outside the sample simply are not part of the denominator.
The practical consequence is that the number belongs in a report with its scope written next to it, every time. A Citation Rate quoted without the words Claude and GPT attached is a number that will be misread by whoever reads the slide six weeks later.
AI Operator Activity is the view with the least hype attached
Of the six reports, AI Operator Activity is the one that describes observable server behaviour rather than sampled answers. Cloudflare's description, as reported on 6 August 2026, is crawl and referral traffic per operator, plus error patterns.
What that combination is good for, as analysis rather than a Cloudflare claim, is the ratio between the two halves. Heavy crawling with almost no referral traffic from the same operator is a fact worth knowing, and it is the sort of thing an ordinary analytics report will not surface because assistant referrals arrive in small, badly labelled volumes. The error pattern half is also concrete: if an operator is repeatedly hitting timeouts or blocked responses on a site, that is a fixable problem with a known owner, unlike a citation score.
The Agent Readiness companion tool
Cloudflare's announcement also covers Agent Readiness, a diagnostic tool that runs checks in four groups. Per the 6 August 2026 report, these are quick wins, which include robots.txt, XML sitemap, AI crawler rules and markdown delivery; technical groundwork, which includes Content Signals, an API catalogue, link headers and agent login; advanced integration, which includes OAuth discovery, MCP, A2A agent cards, a skills index, Web Bot Auth and WebMCP; and commerce, covering x402, ACP and AP2, which the report describes as informational and excluded from the score.
The split between the first group and the third is the useful part for most sites. The quick wins group is made of things a marketing team can verify today without engineering support. The advanced integration group is engineering work with a longer horizon and a much weaker link to any current outcome.
What Cloudflare did not say
The gaps in this announcement are as load bearing as its contents, and several of them are the questions a buyer would ask first.
- No price is published.
- No general availability date has been announced. The product is early access by request from the Overview tab of the Cloudflare dashboard.
- No methodology is published for how answers are sampled: the report does not state how many prompts are run, how they are chosen, how often they refresh, or how a competitor set is defined.
- No stated coverage of Thai language prompts, or of any specific language.
- No stated position on whether the scores are comparable between two different sites, or only meaningful as a trend within one site.
The methodology gap is the one that limits how the numbers should be used. Without a published prompt set and refresh cadence, a movement in Citation Rate cannot be separated from a change in what was sampled that week. That is not an accusation, it is what unpublished methodology means in any measurement product.
What to do while it is early access
The dashboard is requested from the Overview tab of the Cloudflare dashboard, and a request is not a rollout. In the meantime, the measurement it promises can be approximated well enough to be useful.
- Build a fixed prompt set of 20 to 50 questions a real buyer would ask, and freeze the wording so month to month movement means something.
- Run that set across more assistants than the dashboard covers, including at least one it does not sample, and record whether the brand was named, cited, or absent.
- Check server logs or your CDN's bot analytics for AI crawler user agents, and put that beside referral sessions from assistant domains in your analytics.
- Record the scope next to every number you report, so a Claude and GPT figure is never read as an all assistant figure.
- When early access opens, treat the first month as a baseline rather than a verdict, since a first reading has nothing to be compared with.
Most of that work is the same groundwork that supports generative engine optimisation and AI SEO generally, so it holds its value whether or not the product ships broadly.
What this means for Thai marketers
PPC Land's report says nothing about Thailand, the Thai language, or Southeast Asian coverage. There is no Thai detail in the announcement, and none should be read into it.
The reasoning below is inference, not a source claim. A visibility product that does not publish its prompt methodology also does not publish its language mix, so a Thai brand has no stated basis for assuming Thai language prompts are sampled at all. A brand whose customers ask questions in Thai could therefore be scored on English prompts about its category, which is a different question with a different competitor set. The honest position for a Thai marketer is that the coverage is unknown until Cloudflare says otherwise, and that a locally run Thai language prompt set remains the only measurement of Thai visibility a brand actually controls. That is worth pairing with existing ChatGPT SEO work rather than replacing it.
FAQ
Can I use the AEO Visibility Dashboard today?
Only by request, since Cloudflare launched it in early access on 6 August 2026. Access is requested from the Overview tab of the Cloudflare dashboard, and no general availability date has been announced.
What does it cost?
No price has been published. The announcement does not state whether it will be bundled with an existing Cloudflare plan or sold separately.
Does it cover Gemini or Perplexity?
No. Coverage is limited to Anthropic's Claude and OpenAI's GPT models, and Gemini, Perplexity, Copilot and Grok are explicitly outside it. A brand that is invisible in those systems would see nothing about it in this product.
Does it measure Thai language prompts?
The source does not say. No language coverage is stated anywhere in the announcement, and no methodology has been published for how prompts are chosen, so a Thai brand cannot assume Thai prompts are included.
Do I need to do anything right now?
No action is required, since the product is not generally available. The preparation that pays off either way is a frozen prompt set measured monthly and a habit of comparing AI crawler volume against referral traffic in your own logs.
The short version
Cloudflare has built a defensible set of metrics and attached them to half of the assistant market, and the announcement is honest enough about that boundary that anyone quoting the score should be too. Pricing, availability and sampling methodology are all still unpublished, which limits how far a first reading can be pushed. If you want help setting a prompt baseline now, so the dashboard has something to be compared against when access opens, Relevant Audience can help you put one together.
Reporting by PPC Land, published 6 August 2026.
