Our AI Visibility Baseline: 218 Answers, 30 Days

What did 30 days of tracking show?

RankX AI appeared in 44 of the 218 assistant answers analysed in the 30 days to 21 August 2026, an overall mention rate of 20.2 percent. The spread by platform is the finding: Grok, ChatGPT and Gemini all sit near 23 percent, Claude close behind at 20.9, and Perplexity at half the field on 11.4 percent.

These are RankX AI's own numbers for rankxai.com, published as a baseline rather than a boast. The tracking project is eleven days old, the panel is small, and the point of putting the figures on the record is that next month's numbers have something to be honestly compared against.

How was this measured?

Four tracked prompts run nightly against five assistants: ChatGPT, Claude, Gemini, Grok and Perplexity. Each stored answer is analysed for whether it names the RankX brand, and the mention rate divides mentions by answers with a known verdict, so an unknown is never counted as a miss. The window is the trailing 30 days to 21 August 2026; tracking began on 10 August, so it holds eleven nights of runs.

• which tools track how brands appear in ChatGPT results (ready to buy)
• top SEO rank trackers for agencies UK (ready to buy)
• which SEO tools use MCP for AI search tracking (comparing)
• What is RankX and who is it for? (researching)

Google AI Overviews are measured separately, through the rank tracking pipeline rather than the prompt runs: five tracked keywords, checked twelve times each in the window, with the Overview's presence, any brand mention and any citation of rankxai.com recorded per check. The two pipelines are reported apart because they sample differently and answer differently.

Single prompt runs are statistical noise: SparkToro and Gumshoe measured under a 1 in 100 chance that two runs of the same prompt return the same brand list. Everything here is therefore an aggregate over repeated runs, and even so, a four prompt panel is a baseline, not a benchmark.

Why is Perplexity half the rate of the others?

Perplexity named the brand in 5 of 44 analysed answers, 11.4 percent, against 20.9 to 23.3 percent everywhere else. The data cannot say why, and this study does not guess; the gap itself is the finding. It matches the wider evidence that AI platforms are not one channel: in the largest published comparison, only around 11 percent of domains are cited by both ChatGPT and Perplexity.

Perplexity retrieves from its own index, rebuilt in house in 2025, and reranks with cross encoders, so a young brand can be present in one platform's sources and absent from another's. RankX AI reports the five platforms separately for exactly this reason: an average across five different pipelines would have hidden the one number most worth acting on.

Do AI Overviews cite a brand they mention?

Not here, and the split is the sharpest finding in the data. An AI Overview appeared on 59 of the 60 checks across five tracked keywords, 98 percent of the time. The brand was mentioned inside an Overview 12 times, and every one of those mentions came from the brand name keyword; on all four category keywords the count is zero. rankxai.com was cited as a source zero times in the window.

A mention is not a citation: Google will say a name back on a branded query without ever using the site as a source, and citation is the surface where authority is actually earned. Category queries are won elsewhere first, on the third party pages an Overview already trusts, which is the pattern the wider citation research keeps finding. The method behind these definitions is set out in How to Measure AI Search Visibility.

What can a panel this small honestly claim?

Less than a bigger one, and saying so is part of the method. Four prompts and five keywords put wide error bars on every rate here: one answer either way moves a platform by roughly two points, so the per platform ordering reads as rough bands, not a ranking. What a small panel does support is a dated, repeatable series on a fixed method.

This study re-runs monthly on the same panel. The refresh date is stored on the post itself, each update will keep the old figures beside the new ones, and the Last updated line will move only when the data does. If the series stops improving, that stays on the record too.

The vocabulary behind every figure is defined in the glossary, from AI share of voice to AI citation, and the panel design follows the measurement pillar's method: aggregate share over repeated runs, per platform, never a single prompt check.