Who compiles this data
Every dataset is compiled, cleaned and analysed by Kuldeep Paul, data aggregator and data scientist. I'm Head of Marketing at Maxim AI and work on AI search, generative engine optimization and growth. Questions and corrections: mail@kuldeep.cc.
Two kinds of data
Primary data is collected directly. The robots.txt study fetched 10,000 sites myself on Oct 29, 2026.
Aggregated data brings together figures that are scattered across press coverage, official documentation, earnings calls, agency case studies and vendor reports. That covers the ChatGPT Ads benchmarks, advertisers and timeline, the AI engine reference and the user numbers. The value is in the coverage, the deduplication and the links back to sources, not in any single figure.
Collection
- Systematic search. For each dataset I search public reporting, official documentation and practitioner write-ups across metrics, industries and regions, focusing on recent publications.
- Structured extraction. Each result page is read against a fixed schema (for example metric, value, who measured it, sample and period) so every figure arrives with its context.
- Normalisation. Values are parsed to numbers and currencies, metric names mapped to a standard set, and obvious mis-extractions removed: comparison figures for other ad platforms, relative lifts reported as rates, and out-of-range values.
- Deduplication. Identical figures reported by several outlets collapse to one row. The row keeps the most authoritative source (official, then trade press, then agency or vendor) and records how many independent domains repeated it. For the benchmarks, 281 figure mentions reduced to 184 distinct data points.
- Review. Every timeline event, engine profile and crawler entry is written by hand from official documentation where it exists. Advertisers are kept only when a source says the brand ran ads, not merely that it was mentioned or recommended by ChatGPT.
Summary statistics
Headline benchmark figures are medians of distinct reported values. Ranges count at their midpoint; ranges wider than 3× are excluded from medians; cost metrics use USD figures only; CTRs above 5% are treated as outliers. Medians are used because single tests vary widely and the mean is pulled by small, unusual campaigns.
Limitations
- Aggregated figures are only as good as their sources. Many are single-account tests with small budgets, and OpenAI does not publish official benchmarks.
- Public reporting over-represents large brands and English-language markets.
- robots.txt shows intent, not enforcement. Sites also block bots at the firewall, and some bots ignore robots.txt.
- Usage numbers from different companies measure different things (weekly vs monthly, users vs visits).
Updates
Datasets are refreshed on a rolling basis, and each page shows its last update. Every page shows the data that is in its CSV/JSON download, generated from the same file.
Licence and citation
All datasets are published under Creative Commons Attribution 4.0. You can copy, chart, republish and build on them commercially. Please credit Kuldeep Paul, kuldeep.cc and link to the dataset page you used. Each page has a ready-made citation in APA, MLA, Chicago, BibTeX and HTML. The crawler registry incorporates the MIT-licensed ai.robots.txt project.
Citation rules
- Credit Kuldeep Paul, kuldeep.cc wherever you use a figure, chart or table.
- Link to the dataset page, not the CSV or JSON file, so readers can trace every number.
- State the date of the data you used (“as of October 2026”); these figures change.
- If you recalculate, filter or re-chart the data, say it is your analysis of this data.
- When quoting a single reported figure (one agency's CTR, one company's user count), cite its original source too; every row links to it.
- Don't imply endorsement by me or by any company named in the data.
Example: “The median reported ChatGPT Ads CPC is $4.00, according to Kuldeep Paul's ChatGPT Ads Index (kuldeep.cc, October 2026).”