'/%3e%3crect%20x='28'%20y='24'%20width='40'%20height='8'%20rx='2.5'%20fill='%23fff'/%3e%3crect%20x='42'%20y='30'%20width='12'%20height='32'%20rx='2.5'%20fill='%23fff'/%3e%3cpath%20d='M30%2060%20L46%2078%20L82%2034'%20fill='none'%20stroke='%23fff'%20stroke-width='11'%20stroke-linecap='round'%20stroke-linejoin='round'/%3e%3c/svg%3e){kind=small}
Initialed catches 31% more risky clauses than ChatGPT on long contracts
Head to head on the same commercial contracts, scored the same way. On documents over 30 pages — where the expensive clauses hide — the lead reaches 31%. Across all 30 contracts tested, 19% more.
Risky clauses caught
19% more302 of 341 clauses vs 253 of 341
High-severity clauses caught
14% more107 of 124 clauses vs 94 of 124
Contracts over 30 pages
31% moreacross the 8 longest contracts tested
Contracts where it caught more
4 tiedout of 30 contracts
| Head to head, same 30 contracts | Initialed | ChatGPT | Our lead |
|---|---|---|---|
| Risky clauses caught302 of 341 clauses vs 253 of 341 | 88.6% | 74.2% | 19% more |
| High-severity clauses caught107 of 124 clauses vs 94 of 124 | 86.3% | 75.8% | 14% more |
| Contracts over 30 pagesacross the 8 longest contracts tested | 97.4% | 74.6% | 31% more |
| Contracts where it caught moreout of 30 contracts | 21 | 5 | 4 tied |
The longer the contract, the wider the gap
Initialed reviews your contract section by section, so it keeps up as documents get longer. A single chat prompt does not — it caught roughly the same share whether the contract was 5 pages or 50. On the longest contracts we tested, Initialed found 31% more.
Try it on a contract you actually care about
Your first review is free — no credit card, no plugin, any file.
How these numbers were measured
Measured August 2026 using Initialed’s production review pipeline against OpenAI’s then-current flagship model as the ChatGPT baseline, over 30 commercial contracts from the Contract Understanding Atticus Dataset (CUAD v1, The Atticus Project), used under CC BY 4.0 — a public research dataset in which lawyers annotated the clauses that matter. The answer key was written by those annotators before we ran anything, not chosen by us afterwards.
Percentages are clause-level: clauses found divided by clauses the annotators marked, pooled across all 30 contracts. “19% more” and “31% more” are relative — 302 clauses found against the baseline’s 253, and 97.4% against 74.6% on the eight contracts over 30 pages. A clause counts as found only when the quote the review points to overlaps the annotated clause; the same scoring code and threshold scored both sides. The baseline received the whole contract, the side being represented, and the same red-flag checklist Initialed uses — more guidance than a typical user provides. No contract was shortened to fit.
Scope and limits: this measures clause detection, not how often either tool raises an issue that turns out not to matter — we do not yet publish a figure for that, and Initialed reports substantially more findings per contract. CUAD records which clauses appear in a contract; which of those count as risky depends on which side you are on, and that judgment is ours. Both products change over time and neither returns identical output twice, so results will vary. Nothing here is legal advice or a guarantee about any individual contract.
ChatGPT and GPT are trademarks of OpenAI. Initialed is not affiliated with, endorsed by, or sponsored by OpenAI. The comparison reflects a single prompt sent to the model named above via its public API.
We use strictly necessary cookies to run the app. With your consent, we also use analytics and marketing cookies to improve and promote our services. See our Privacy Policy.