# AI search visibility repeat video transcript

Study date: 10 November 2026

## What we did

We repeated the same four commercial SEO prompts across five AI systems and five planned runs. The wording stayed fixed so we could compare what changed between valid answers. The workbook records two discrete rounds on 15 October and 10 November 2026, 26 days apart. This was not continuous monitoring. The protocol planned 100 observations. Eighty-eight were valid. Twelve Claude runs were not observed after the free-account quota expired. Those missing runs are a protocol limitation, not negative results.

## Why we did it

One AI answer can look like a market position. We wanted to know whether the next valid answer would agree, and what that changed commercially. One answer is not a baseline.

## Why it matters

The absence repeated. The recommendations did not. Across 60 like-for-like reporting slots, none reproduced the same ordered supplier list. Nine retained the same first supplier. Fifteen of 220 comparable list positions contained the same supplier.

For one prompt, ChatGPT Search run one recommended Re:signal, Claneo, Orainti and Delante first. Run two recommended Anicca, Distinctly, The SEO Works and HECHT INS GEFECHT first. These are generated supplier suggestions, not a stable market ranking.

That creates a commercial risk. One answer is a photograph of moving traffic. It is not a market map. A business can commission content, accept a visibility claim or commit budget before it knows whether the observation repeats or why it occurred.

## What was the outcome

letsrocc appeared in none of the 88 valid responses and received no citations. The October clean baseline was zero out of sixty. The November like-for-like repeat across ChatGPT Search, Google AI Mode and Gemini was also zero out of sixty. No directional improvement was observed in the tested prompt set.

The result is useful, but the cause remains unknown. The study cannot establish whether content, links, schema, llms.txt, geography, retrieval or model behaviour caused the result.

## Next step

A bounded AI evidence question can sit inside the Marketing MRI when the decision crosses marketing, measurement and commercial outcomes. The work tests the claim before the business funds the remedy. It records the situation, repeats the observation, examines recurrence, citations, sources and page proof, then separates justified action from what should wait.

The output is a decision-grade baseline: what appears reliably, what changes between runs, which sources support it, what remains unproven and what to test next.

The article connects that evidence to the Marketing MRI, its scope, costs and next step.

## Who stands behind the work

letsrocc is led by Chris Wheeler, Chief Systems Operator with 31 years in digital, and Claudia Misselwitz, with 21 years in SEO strategy and governance. We report what was observed, state the limits and do not turn generated supplier suggestions into rankings.

## Continue with the evidence

Read the study and method at https://letsrocc.com/en/insights/ai-search-visibility-repeatability-study.

Watch Chris and Claudia at https://www.youtube.com/@letsrocc.

No visibility guarantees. No manufactured score.