ai-visibility-daily-record

A dated, public record of which businesses AI engines name when people ask for a recommendation. Each row in the answers table is one question on one date to one engine. Every claim owes the future a receipt.

Edition 1 is final. Data through 2026-10-13.

Counts: 73 questions, each asked in the same words on every run day it was in the roster (55 market or category questions and 18 questions asked to see whether one named business comes up); 25 US local markets plus nationwide questions (27 market labels as written in the roster); 37 niche labels as written in the roster; API engines asked: anthropic, gemini, openai, perplexity, plus Google results; 42 run days from 2026-09-01 to 2026-10-13.

The record is kept by Namebeam, a Carter Enterprise LLC company.

What is in this repository

Everything is under edition-1/. Edition 1 covers 2026-09-01 to 2026-10-13 (UTC dates): 42 run days, 12,180 answer rows: 8,160 engine rows (6,012 answered with at least one name, 318 answered with no name, 1,830 failed calls kept as rows with their error), 2,010 Google result rows (1,118 extracted, 892 extraction failed) and 2,010 engine agreement rows. The default config adds a run_status column so a failed call is never read as a zero.

The method note is edition-1/README.md. The license is CC BY 4.0; the full text is edition-1/LICENSE. The code that computes the metrics is in edition-1/method/. edition-1/MANIFEST.sha256 lists the hash of every file, and edition-1/verify.py checks them.

How the data is collected

1. Each day the same fixed questions are asked of Perplexity, OpenAI, Anthropic and Gemini through their APIs, and the Google results page for the same question is read.
2. Every answer is saved as a raw file in the public GitHub repository, and the answers table names that file and carries its SHA-256.
3. The record is append only. A day on which an engine failed is recorded as failed, not as zero. Losses print at the same size as wins.

Perplexity runs with web search on (Sonar chat completions to 2026-09-27, the Agent API with a forced web search from 2026-09-28). OpenAI, Anthropic and Gemini run with no search tool, so they answer from the model alone. The engines are not like for like.

The answers table

Known limits

• These are API calls, not the apps people open on a phone. A person using an app may see something different.
• The model that answered is not recorded in the rows.
• Names are raw strings found by a program. The metrics give each figure twice, raw and after a rule-based name gate, with the denominator beside it.
• OpenAI deprecated the gpt-5-mini snapshot (notified 2026-06-11, removed from the API 2026-12-11, replacement gpt-5.6-terra; OpenAI deprecations page, read 2026-10-08). The collector still reads gpt-5-mini. A model change will be dated in the method note before the first changed row.
• Gemini returned some answers until 2026-09-26 and none from 2026-09-27 (HTTP 402, prepayment credits depleted). Its figures are small and carry their denominators.
• The metrics read the 38 market and niche cells asked on or before 2026-09-27. Research segments added after that date are in the answers table and not in the metrics.
• The window after 2026-09-27 is short, and the Perplexity method changed on that date. A change between the two windows mixes the method change with change over time.
• The Siri rows are typed or spoken by hand on one phone. They are not a sample of Siri users.
• A count is a count. It shows what was returned on the days an engine answered. It is not a rating or a ranking.

Examples

Perplexity on Sonar, 2026-09-01 to 2026-09-27: 2,398 of 5,486 business names (43.7 percent) returned on one day were returned again the next day. Raw names, next day answered, 38 market and niche cells.

In the same window the overlap between two consecutive days' name lists had a median of 30 percent over 832 day pairs. Raw names, both days named at least one business.

Load the data

from datasets import load_dataset

ds = load_dataset("Namebeam/ai-visibility-daily-record", "answers", split="train")
print(ds[0])

Check the files

git clone https://github.com/mrcrtr1979-droid/namebeam-visibility-index
python3 namebeam-visibility-index/releases/edition-1/verify.py --rebuild --metrics

The last line reads VERIFY MATCH when every file agrees with the manifest and the tables rebuild from the raw files.

Citation

@dataset{namebeam_ai_visibility_daily_record,
  author    = {{Carter Enterprise LLC}},
  title     = {Namebeam AI Visibility Daily Record},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/Namebeam/ai-visibility-daily-record}
}

License

CC BY 4.0. Credit Namebeam and link the record.

Links

• Namebeam: https://namebeam.ai/
• Public record on GitHub (append only, losses included): https://github.com/mrcrtr1979-droid/namebeam-visibility-index
• Sibling dataset with the daily CSV mirror: https://huggingface.co/datasets/Namebeam/namebeam-e1-ai-referral-index

Maintainer

Carter Enterprise LLC, 30 N Gould St, Ste 65270, Sheridan, WY 82801. Questions about the record: namebeam.ai.