Does an AI shortlist show what buyers actually purchase?
An AI shortlist does not show what buyers actually purchased. The September Index retained 2,095 valid answers from 2,100 scheduled observations captured 22 and 23 September 2026, but those were generated responses to a fixed research bank. They were not 2,095 buyer journeys or transaction records.
What we measured
Measured. The September Index method covered 14 categories and 286 checked vendors. Each category had ten buyer questions asked three times on ChatGPT, Claude, Perplexity, Google AI Overviews and Google AI Mode. Five scheduled observations were excluded, leaving 2,095 valid answers; release 2026-09 was published 23 September 2026.
The unit is an answer observation. A vendor can appear in several runs, questions or engine responses, and the same question is intentionally repeated. None of those repetitions records another person buying the product.
The dataset provides no purchase count, conversion rate or measured market share.
Every figure comes from the published method, v1.1.
What this means for you
Reasoned. Use a shortlist as evidence about the options an engine put before a question, then keep any buyer action in a separate record. A person may investigate a named product, reject every option or never make a purchase. The generated list alone cannot distinguish those outcomes.
Even a suitable recommendation would not establish a transaction. Price, requirements, procurement and the person's own evaluation remain separate parts of a buying decision. They need evidence appropriate to that decision rather than a renamed visibility percentage.
- If you sell software: report answer appearances as answer appearances. Do not turn a naming count into demand or sales without the corresponding records.
- If you buy software: treat the returned options as leads for investigation, then verify fit with your actual requirements.
- If you report business outcomes: preserve visits, buyer self-reports, opportunities and sales as distinct events with their own dates and definitions.
A later comparison can connect those records only when a stated join is supported. Where a link is missing, leave the contribution unknown. An answer sample can still be useful without being presented as a substitute for evidence of what people did.
Also asked as
Does an AI shortlist show what buyers actually purchase?
Related
- September Index data
- AI software shortlists
- Business evidence standard
- Connect visibility with pipeline
Next step
Keep the output of Free 10-question check (one engine) with your answer observations; it does not add a buyer or sale to a business ledger. No email is needed; runs are limited per network and per day, and by the site's daily allowance.
When an answer exposes a particular product fact that needs correction, Fix one answer, $490. Assess any later purchase outcome using its own evidence.
Prepared by the Broadcastwell team.
Page owner: Priya Nair, Management Consultant. Corrections: hello@broadcastwell.com