Evidence annex
Nine runs of HubSpot’s AI Search Grader, in full
This is the run log behind the article An AI visibility score for a brand that does not exist yet. Nine graded tests on the brand Syndral, run between 21 and 23 July 2026, one variable changed at a time, with the exact inputs, the scores that came back, and the archived artefacts for each run that has them.
OpenAI, Perplexity and Gemini are the overall score out of 100 from each engine, not the sentiment sub-scores quoted below. Scroll the table sideways on a narrow screen. Emerald marks the field that changed from the previous run. All times UTC+0700.
| # | Run | Date and time | Product field | Industry field | Changed from previous run | OpenAI | Perplexity | Gemini | Archived artefacts |
|---|---|---|---|---|---|---|---|---|---|
| 01 | Test 1 | 21 Jul 2026 · ~19:20 | “None” | “Agency” | The original submission | 33 | 7 | 30 | No artefact. Reproduced as run 08. |
| 02 | Test 2 | 21 Jul 2026 · ~20:10 | “Ai Visibility Agency” | “Service Agency” | Both free-text fields rewritten | 36 | 32 | 40 | No artefact. Reproduced as run 03. |
| 03 | Baseline 1 | 23 Jul 2026 · 14:53:11 | “Ai Visibility Agency” | “Service Agency” | Nothing. Test 2 resubmitted | 36 | 32 | 40 | |
| 04 | Baseline 2 | 23 Jul 2026 · 15:02:29 | “Ai Visibility Agency” | “Service Agency” | Nothing. Test 2 resubmitted | 36 | 32 | 40 | |
| 05 | Baseline 3 | 23 Jul 2026 · 15:05:29 | “Ai Visibility Agency” | “Service Agency” | Nothing. Test 2 resubmitted | 36 | 32 | 40 | |
| 06 | Capitalisation | 23 Jul 2026 · 15:10:53 | “AI Visibility Agency” | “Service Agency” | One letter, “Ai” to “AI” | 36 | 32 | 40 | |
| 07 | Industry word | 23 Jul 2026 · 15:17:11 | “Ai Visibility Agency” | “Professional Services” | One word, “Service Agency” to “Professional Services” | 36 | 28 | 45 | |
| 08 | Test 1 replay | 23 Jul 2026 · 15:23:30 | “None” | “Agency” | Back to Test 1’s exact inputs, two days later | 33 | 7 | 30 | |
| 09 | URL only | 23 Jul 2026 · 15:35:12 | As run 07, unchanged | “Marketing Agency” | The form step itself, and the industry word with it. A results address loaded directly, no form and no click. Two things move here rather than one, because the run has to use a combination the form never sent. | 36 | 36 | 41 |
Local time, UTC+0700. Runs 01 and 02 predate the protocol, so the log records them to the nearest ten minutes.
HubSpot renamed the tool to AI Search Grader after these runs. The archived reports carry its name at the time, AEO Grader. They have not been altered.
The two comparisons the protocol was built for
Runs 03 to 06 · identical or near-identical inputs
Perplexity overall. Identical to each other and to Test 2, two days earlier, down to the written analysis on runs 03 to 05. Run 06 carries a filled form rather than a result summary, so its score is logged and its analysis is not.
Runs 03, 07, 09 · one word in the industry field
Perplexity overall, an 8-point spread. The Gemini sentiment totals moved too: 65/100 for “Service Agency”, 78/100 for “Professional Services”, 68/100 for “Marketing Agency”.
The third wording, “Marketing Agency”, comes from run 09, loaded directly by URL. The word travels as a URL parameter either way: the form only builds that address, which is what run 09 demonstrates.
Tool: HubSpot AI Search Grader, free and public. Every run here can be reproduced in minutes.