Your baseline, read back as a report
Every run you logged in the Baseline Kit — scored per platform, grouped by topic, ranked against whoever got named instead, and turned into a fix list. Nothing is fetched and nothing is sent. This page only reads what is already in your browser.
A filled report, so you can see what yours will look like
This is the locked baseline of Eggleston Office, an office furniture dealer in Dallas–Fort Worth, the founder’s family business, and Command Module’s own first customer: 100 buyer questions put to seven engines, three passes over, locked on 2026-09-19. Every figure is measured. Nothing is projected, and nothing is blended. Your own runs replace this section the moment you log one.
Per-engine visibility
Pass one of three. Each bar divides by the answers that engine gave: a question it had no answer to, or a leg that did not run, is counted beside the bar and kept out of it.
9.4% is the average of those bars and it describes none of them. Per engine, always: the same rule this page applies to your own runs.
ChatGPT
20%
Named in 20 of 100 answers.
- Named first
- 14 of 20
- Link rate
- 95% 19 of 20
- Named again
- 41.9% 26 of 62
- Same verdict
- 82% 164 of 200
Claude
0%
Named in 0 of 92 answers (a floor: answers measured before 2 October 2026 were cut short). Of its 100 questions, 3 drew no answer to judge and 5 did not run; neither counts.
- Named first
- —
- Link rate
- —
- Named again
- —
- Same verdict
- —
Never named on this engine in pass one, so first place, link rate and named again have no denominator. Same verdict would only count the passes agreeing on a miss. Each reads “—”.
Perplexity
10%
Named in 10 of 100 answers.
- Named first
- 6 of 10
- Link rate
- 0% 0 of 10
- Named again
- 53.6% 15 of 28
- Same verdict
- 93.5% 187 of 200
Gemini
10%
Named in 10 of 100 answers.
- Named first
- 6 of 10
- Link rate
- Not recordedread and cited kept together
- Named again
- 51.9% 14 of 27
- Same verdict
- 93.5% 187 of 200
Link rate leaves out all 10 answers that named the business. This engine listed the pages it read in the same list as the ones it cited, and this pass was run before the two were kept apart, so whether those answers linked the site cannot be told.
Grok
21%
Named in 21 of 100 answers.
- Named first
- 14 of 21
- Link rate
- 90.5% 19 of 21
- Named again
- 66% 33 of 50
- Same verdict
- 91.5% 183 of 200
Google AI Overviews
2.9%
Named in 2 of 70 answers. Of its 100 questions, 30 drew no answer to judge; neither counts.
- Named first
- 2 of 2
- Link rate
- 0% 0 of 2
- Named again
- 100% 3 of 3
- Same verdict
- 100% 115 of 115
Every rate on this card divides by 2, so read each one as a count.
Google AI Mode
2%
Named in 2 of 100 answers.
- Named first
- 2 of 2
- Link rate
- Not readablecards link via Google
- Named again
- 100% 4 of 4
- Same verdict
- 100% 200 of 200
Every rate on this card divides by 2, so read each one as a count.
Each square is one question, in order: lit when named, dark when missed, outlined when the engine gave no answer to judge, dashed when the leg did not run.
Who gets named instead
The five tracked DFW dealers, counted across the 662 answers in pass one. Eggleston finishes fifth of five on a 10.3% share of voice: 65 of the 631 times any of the five was named.
Questions lost on every engine
62 of 100 questions named Eggleston on none of the engines that answered them in pass one. Every question here is graded winnable: Eggleston could truthfully be the answer. The ten below drew the most tracked dealers. Claude’s answers were measured before 2 October 2026 and cut off part way, so a question here may have named Eggleston further down.
+ 52 more in the raw sample JSON
Where the engines look
Share is a kind of site’s cut of every source listed, so the column adds to 100. Coverage is how many of the 575 answers that listed a source listed that kind at least once, so it can add past 100. A listed source is a page the engine read or cited.
Visibility by topic
Pass one, each topic over the answers it drew across the seven engines, lowest first.
Where this came from. eggleston-audit/baselines/baseline-20260919-153713.json, schema 2, locked 2026-09-19, append-only and hashed, read with its three pass files. Pass one ran 2026-09-18. Read the raw JSON and check every number on this page against it.
No baseline in this browser yet
The Scorecard reads the runs you log. It needs the prompt set and the scores you record in the Baseline Kit — generate a set, click the platform links, tick what you see, and come back.
Rather not spend the week clicking? We’ll run 2,100 of them for $149 and send back this same page, filled in.
01 · What you told the Kit
Everything below is computed from this, and only this.
02 · Per-platform visibility
Three separate numbers, deliberately. A blended "AI visibility score" averages away the only thing you can act on — engines disagree, and the fix differs per engine.
03 · Visibility by topic
Your prompt categories, each counted across every run you logged. Winnable marks the ones where a business your size can realistically get named.
04 · Who gets named instead
Tallied from the "named first instead" box, counted once per question. This is the leaderboard your buyers actually see.
Prompts you lose
A run where an engine named someone and did not name you. Each one is a buyer question answered with a competitor.
05 · Sources the engines cited
From the "cited sites" box. These are the pages standing between a buyer's question and an answer that is not you — and the shortest route onto them is usually one real, useful contribution.
06 · Fix list, ranked
Derived from your zero-coverage topics, the competitor ahead of you, and the domains the engines already trust. Ranked by how many of your logged prompts each one touches.
Take it with you
The report is yours. Nothing here was uploaded, so nothing here can be deleted by us.
It will not tell you a number it did not measure
If you logged eleven runs, it reports eleven runs. It will not estimate the two engines you skipped, it will not project a score for an engine with no logged-out page, and it will not blur a fabricated figure behind a paywall and call it a preview. Every number on this page traces to a row you ticked yourself — which is also why the honest answer is sometimes "not enough runs yet."
That last one is the catch, and it is worth saying plainly: the thin part of a DIY baseline is the run count. The method here is the one we sell, unchanged. But 100 questions across seven engines is 700 runs before you repeat anything, and a second and third pass takes it to 2,100 — and three passes is the smallest number that separates "the engine doesn't know us" from "that one was a fluke." Almost nobody gets past the first pass by hand, which is the only reason the rung below exists.