Which AI ad tools ban benchmarks and comparisons?

What is a benchmarking clause?

A benchmarking clause bars users from testing a service against its rivals, or from publishing how it performs. Database software had them first. Oracle set up the original “DeWitt clause” after a 1982 benchmark by David DeWitt displeased Larry Ellison. Such a clause forbids publishing benchmarks the vendor has not sanctioned (Wikipedia).

The wordings differ in what they bar. Luma’s bars publishing results.

HeyGen’s bars accessing the services “for benchmarking, comparative or competitive purposes”.

Canva’s and Motion’s bar access for benchmarking too, Motion’s in capitals before section 1.

AdCreative.ai, The Brief, Pencil and Claid.ai bar using or analysing the service for it, and Jasper bars monitoring for it.

Of the 14, only Meta’s names a tool’s outputs.

Which tools have no such clause?

In 61 tools’ terms read on 12 October 2026 we found no clause that bars benchmarking or comparison. We searched each document for benchmark, competitive, comparative, performance information, publish results, reverse engineer and compete.

Twenty-four of them bar using the service, or its outputs, to build or run something that competes with it. None of those clauses names benchmarks or comparisons.

Higgsfield’s is typical: “(iii) use or access the Service or any Outputs to develop, modify, fine-tune, distill, or improve any products or services that compete with the Service” (Terms of Use, 26 July 2026).

Kling and ElevenLabs use the word benchmarking only for their own analysis of usage data.

Terms not readable

TikTok Symphony’s pages say its GenAI tools fall under “TikTok’s Creative GenAI terms”. No public URL for them turned up on 12 October 2026, and the ads policy address returned a login page.

Magnific’s general Terms of Use, dated June 2026, carry no benchmark clause. They now point to separate API Services and AI Products terms, and both pages answered 403.

On 24 September 2026 the API clause read: “Any such use for competitive, benchmarking, or replicative purposes shall require Magnific’s prior written authorization in the form of a specific license”. It reached API use only.

What changed since September 2026?

Every clause in the table reads word for word as it did on 23 and 24 September 2026. Five documents carry new dates and no new clause. Google Cloud’s terms date from 8 October 2026, with section 7 unchanged. Arcads dates its version 8 October, ElevenLabs 9 October, Maxfusion 28 September and Hedra 4 September.

Ten tools joined the register in this read, Foreplay and Madgicx among them, and none carries a clause.

Meta announces a change dated 30 October 2026. The 5 October 2026 capture carries the banner “We’re updating the Self-Serve Ad Terms to incorporate the Ad Creative Generative AI Terms. These updates will go into effect on October 30, 2026.”

The preview Self-Serve Ad Terms, effective 30 October 2026, do not contain the “outside of Meta’s platforms” sentence. Their AI paragraph bars using the features or their outputs “to develop models, algorithms, or systems to compete with Meta”.

What this means for published comparisons

Each clause names who it binds. HeyGen’s list opens “you may not”. Luma’s binds the customer, who “will not permit its Users or anyone else” to do the listed acts. Motion’s sentence covers any access to the service.

Three documents set out what is allowed. Luma’s API Terms of Use, dated 28 April 2026, require “prior written consent” for published benchmarks, “provided that Customer may conduct internal, non-published performance evaluations and security assessments of the APIs”.

Google Cloud’s section 7 lets a Cloud customer publish benchmark results, Veo’s included. The conditions: the disclosure holds all the information needed to replicate the tests, and the customer lets Google benchmark its public products in return.

Jasper’s Usage Policies except “uses that are expressly permitted”, for example in an Order Form.

Vidu’s clause names competitive analysis and no benchmark.

Its terms do not say whether a published side-by-side comparison counts as one. Zeely’s 6.2 does not say whether outputs count as “content or features of the App”.

Every adtestbench tool page synthesises public user reviews and the vendor’s own pricing and terms pages, as our method page sets out. Commercial-use terms by plan sit in the AI commercial use matrix, and the marks each tool leaves on its outputs on the watermarks page.

How was the register built?

A row keeps its tool slug as its anchor, so a link such as /data/benchmark-bans/#luma stays stable. Corrections show in this page’s list of changes and on updates and corrections.

How to cite this data

adtestbench, "Which AI ad tools ban benchmarks and comparisons?", https://adtestbench.com/data/benchmark-bans/, updated 12 October 2026.

• CSV file /data/benchmark-bans.csv