Receipts Index

Namebeam AI Visibility Index

A dated public record of which local businesses four AI engines name when they are asked the same buyer questions. Every answer is saved with its date and a SHA-256 hash of the raw file.

Edition 1

• Window: 2026-09-01 to 2026-10-13, 42 run days (no run on 2026-10-05).
• Questions: 73. Of these, 53 were asked of all four engines and 20 of Perplexity only.
• Calls: 8,160 API calls. 6,330 were answered: 6,012 named at least one business and 318 named none. 1,830 failed and are kept as failed rows.
• Engines: Perplexity with web search on (Sonar chat completions to 2026-09-27, Agent API with a forced web search tool from 2026-09-28). OpenAI, Anthropic and Gemini through their APIs with no search tool switched on.

DOI: 10.5281/zenodo.23626831

Browse the tables: https://markets.namebeam.ai/edition-1/

Hugging Face dataset: https://huggingface.co/datasets/Namebeam/ai-visibility-daily-record

Gaps in Edition 1

• Gemini returned no answers for the last 16 run days. Every Gemini call from 2026-09-27 failed because the prepaid credit ran out.
• OpenAI name lists are empty from 2026-09-01 to 2026-09-05. The raw answer text is present for those days.

One finding

Over the whole Edition 1 window, two engines' same-day lists of businesses overlapped by 6 percent at most (Gemini and Perplexity, 280 same-day comparisons). Every other pair overlapped by 1 percent or less.

Source: releases/edition-1/metrics/metrics_v2_sentences.md, metric A.gated.pair, whole Edition 1 window.

How to verify

Run these from the root of a clone of this repository. The raw answer files are in corpus/e1/.

Check one raw answer file against the hash recorded for it in the answers table:

sha256sum corpus/e1/NB-CZ-API_2026-09-10_atrix_digital_anthropic.json
grep -c bf425f6540799a10f0ea413c4d64c4f0ca56641970099f0fac24fdb694514c75 releases/edition-1/data/answers_2026-09-01_to_2026-10-13.csv

The first command should print bf425f6540799a10f0ea413c4d64c4f0ca56641970099f0fac24fdb694514c75 and the file name. The second should print 1.

Check the answers table itself against MANIFEST.sha256:

cd releases/edition-1
grep 'data/answers_2026-09-01_to_2026-10-13.csv$' MANIFEST.sha256 | sha256sum -c -

Expected output: data/answers_2026-09-01_to_2026-10-13.csv: OK (hash 398f7cc5f76a3060eba318d0231e4b32f92a3155064debb0bc68f0c9fb499f03).

Check everything at once:

python3 releases/edition-1/verify.py

With no flags, verify.py reads MANIFEST.sha256 and PACK.json, hashes every file listed in the manifest (33 files), prints MATCH or MISMATCH for each, and flags any file in releases/edition-1/ that the manifest does not list. The last line reads VERIFY MATCH and the exit code is 0 only when everything agrees. It needs the whole releases/edition-1/ folder and nothing outside the Python standard library.

Two optional flags go further. --rebuild rebuilds the answers table from the raw files in corpus/e1/ and compares it with the manifest. --metrics recomputes the metrics table with the code in releases/edition-1/method/ and compares it.

Licence scope

CC BY 4.0 covers the compilation (selection, arrangement and structure of the tables), the code, annotations, derived columns and documentation. The raw answer text returned by OpenAI, Anthropic, Google Gemini and Perplexity, and the search result rows, are published as dated factual records and are not licensed by Namebeam; provider terms may apply.

Disclosure

Namebeam sells AI visibility checks built on this method.

Namebeam, a Carter Enterprise property. This is the first leg.

Earlier notes (before Edition 1)

Namebeam AI Visibility Index

An append-only, dated, public record of one question: when someone asks an AI engine to recommend a tool that checks AI visibility for a business, does the engine say Namebeam?

This dataset is published by Namebeam (namebeam.ai), an AI visibility service that checks whether ChatGPT, Claude, Perplexity, Gemini, and Google AI name a specific business when a customer asks who to hire. Namebeam runs the same check on itself that it sells, and publishes the results here, win or lose.

The record so far

As of 2026-07-31: five engines run, zero of five named Namebeam. That number is our starting line, not a confession. Every test is dated, every miss is logged, every fix gets re-tested, so the climb gets proven on this record.

Files

• data/ai-visibility-index.json: the corpus. One row per engine per run. Fields: check_id, date_utc, business, niche, market, engine, prompt_variant, run, prompt_text, answer_verbatim, business_mentioned, competitors_mentioned, factor_scores.
• data/heartbeat.log: the weekly sensor log.
• .github/workflows/ai-visibility-sensor.yml: a weekly GitHub Actions heartbeat (Mondays 13:00 UTC) that keeps the record alive and opens a re-check issue.

Rules of the record

1. Append-only. Rows are never edited or deleted.
2. No fabricated rows. A failed run is logged as a failed run.
3. Losses print the same size as wins.
4. Memory-personalized engine sessions are refused; only customer-comparable runs enter the corpus.

The human-readable version

• Full customer-zero ledger with transcripts: https://proof.namebeam.ai
• What Namebeam is: https://namebeam.ai
• Run the same free check on your own business and get a dated receipt: https://check.namebeam.ai

Namebeam scores AI visibility on five factors: Findable, Reachable, Quotable, Bookable, Trusted. Instead of an invented score, the deliverable is the actual dated transcript of what each engine said.

Operated by Carter Enterprise LLC, Sheridan, Wyoming.