Evaluation Stage First: What 19 Tracked Companies Show About AI Visibility
September 10, 2026
Across the 19 companies with a completed scan in our own platform between 10 July and 7 September 2026, every single one appeared in at least one evaluation-stage answer. Only 12 appeared in any discovery-stage answer, and only 4 earned a citation of any kind. The pattern is consistent enough to change what an early-stage team should publish first.
Table of Contents
- The data
- What stands out
- What this implies for an early-stage team
- Method and limits
- FAQ
- Related reading
The data
Sample: the most recent scan for each of the 19 tracked companies in the Monroya database with a completed visibility snapshot, 10 July to 7 September 2026. Prompts vary by company because each set is built around that company's own buyers. This is a small sample and it is our own customer base, so read it as a directional pattern rather than a market statistic.
| Measure | Result across 19 companies |
|---|---|
| Appeared in at least one evaluation-stage answer | 19 of 19 |
| Appeared in at least one decision-stage answer | 16 of 19 |
| Appeared in at least one discovery-stage answer | 12 of 19 |
| Earned any citation | 4 of 19 |
| Average evaluation-stage inclusion rate | 41.6% |
| Average decision-stage inclusion rate | 42.5% |
| Average discovery-stage inclusion rate | 29.5% |
| Average citation rate | 0.8% |
| Highest citation rate observed | 7.4% |
What stands out
1. Evaluation is the universal entry point. It was the only stage where every company in the set had some presence. Discovery, the stage most content plans are built around, was where a third of these companies had no presence at all.
The mechanism is not mysterious. Evaluation questions ("best tools for X", "which vendors should we shortlist") are answered from documents that are already shaped like shortlists: comparison pages, listicles, head-to-head write-ups. Those documents are also the easiest for a small company to get into, because they are published by third parties who need examples. Discovery questions pull from broad authoritative explainers, which is a game weighted heavily toward incumbents.
2. Mentions and citations are not the same problem. Companies averaged around 40% inclusion at the evaluation stage while averaging 0.8% citation. Being named is common. Being linked as the source is rare. Those need separate plans: a mention comes from being described somewhere credible, a citation comes from being the page worth quoting.
3. Discovery presence lags, and that is normal. If you are early and invisible in discovery answers, that is the expected reading, not a failure. It is also the most expensive gap to close, which is a reason to close the cheaper one first.
What this implies for an early-stage team
- Publish the comparison content before the introductory explainer. Honest head-to-heads and category shortlists are the fastest realistic opening, which is consistent with why comparison pages get pulled into shortlist answers.
- Get into other people's lists. A third-party comparison already being cited is worth more than a new page nobody quotes yet.
- Treat citation as a separate, slower goal. Original data and documentation earn citations; marketing pages usually earn mentions.
- Measure by stage, not in aggregate. A single blended score hides exactly the asymmetry that tells you what to do.
Method and limits
Each prompt is run multiple times per assistant across ChatGPT, Claude, Gemini and Perplexity, and inclusion is reported as a rate rather than a yes or no. Inclusion rates are the values recorded on each company's most recent snapshot. Nineteen companies is a small sample drawn from one platform's customers, skewed toward B2B services and software, and prompt sets differ between companies. It does not establish a market-wide rule. It is enough to say that in this set, the evaluation stage was where presence started.
Want your own version of this reading? The free AI visibility check runs a sample of buyer prompts and returns a stage-level score with no signup, and the demo shows the full view built from a real scan.
FAQ
Why is the citation rate so low? Citations require the assistant to credit a specific page as its source. Most brand appearances are descriptive rather than sourced, and assistants preferentially cite third-party and reference material over vendor sites.
Does a 40% inclusion rate mean I appear in 40% of buyer conversations? No. It means the brand was named in roughly 40% of the recorded responses to that company's evaluation-stage prompt set. Real buyers ask different questions in different words.
Is 19 companies enough to draw conclusions from? Not for a market-wide claim. It is a directional signal from a consistent measurement method, and we will publish an updated version as the sample grows.