What is cohere-ai? Cohere's AI crawler
cohere-ai is a user-agent token associated with Cohere, the enterprise AI company. IndexHalo classes it as a training crawler.
67 of the 389 major websites we measure block cohere-ai in robots.txt (17%), making it number 9 of the 30 AI crawlers we track for how often it is blocked. 63 of those sites name it directly in their robots.txt; the rest are covered by a wildcard rule.
Key facts
| Fact | Detail |
|---|---|
| Operator | Cohere |
| Used for | Cohere's models |
| Type | Model training. Collects content that may be used to train AI models. Blocking it does not usually remove a site from AI search. |
| robots.txt token | cohere-ai |
| Operator's documentation | None we could verify |
What blocking cohere-ai does
We did not find a documentation page from Cohere that we could verify, so this page reports only what robots.txt files say about it.
How to allow or block cohere-ai
Add one of these groups to the robots.txt file at the root of your domain. It is a stated preference that well-behaved crawlers read, not an enforcement mechanism.
User-agent: cohere-ai Disallow: /
User-agent: cohere-ai Allow: /
cohere-ai user agent string
We did not find a full string from its operator, so look for cohere-ai in the user agent field of your logs.
How to verify a request is really cohere-ai
Anyone can put a crawler's name in a user agent string, so the name alone proves nothing. We did not find a published IP list or reverse DNS rule from Cohere.
How to whitelist cohere-ai in a firewall or CDN
robots.txt only states a preference. A firewall, CDN or bot protection service can still refuse the request, and the crawler then gets an error even though robots.txt allows it.
- Check that no group in your robots.txt blocks it, using the rule above.
- Add an allow rule in your firewall or CDN that matches the name in the user agent. Keep the rule narrow, because anyone can copy a name.
- Check your server logs for a 200 response from cohere-ai afterwards.
On Cloudflare, also check AI Crawler and Agent access policies, which can block search, agent and training traffic across a site.
Why cohere-ai gets a 403 or 405 error
A 403 means the server refused the request, which usually points to a firewall or bot rule. A 405 means the server refused the request's method, so check any rule that only permits certain methods. A crawler that obeys robots.txt won't request a page it is told to avoid, so an error in your logs points to your server's configuration. To stop it deliberately at the server, this nginx rule returns a 403:
if ($http_user_agent ~* "cohere-ai") {
return 403;
}Which sites recently changed their rules for cohere-ai?
| Date first seen | Site | Change |
|---|---|---|
| 31 August 2026 | fastcompany.com | Started blocking |
| 31 August 2026 | inc.com | Started blocking |
See the full change log.
Who blocks cohere-ai, by sector
| Sector | Sites measured | Blocking | Share |
|---|---|---|---|
| News publishers | 44 | 27 | 61% |
| Technology media | 25 | 14 | 56% |
| Social and media platforms | 28 | 7 | 25% |
| Finance and business media | 29 | 7 | 24% |
| Travel | 19 | 3 | 16% |
| Health publishers | 16 | 2 | 13% |
| Reference and research | 25 | 3 | 12% |
| E-commerce and retail | 19 | 2 | 11% |
| Developer references | 34 | 1 | 3% |
| SaaS platforms | 54 | 1 | 2% |
| AI companies | 28 | 0 | 0% |
| Government and institutions | 19 | 0 | 0% |
| Hosting and infrastructure | 23 | 0 | 0% |
| SEO and marketing tools | 26 | 0 | 0% |
Sites blocking cohere-ai
academia.edu, aljazeera.com, amazon.com, apnews.com, arstechnica.com, asahi.com, barrons.com, bbc.co.uk, bloomberg.com, buzzfeednews.com, cnbc.com, cnn.com, cntraveler.com, disneyplus.com, drugs.com, dw.com, fastcompany.com, figma.com, flickr.com, forbes.com, france24.com, ft.com, geeksforgeeks.org, gizmodo.com, globeandmail.com, hackernoon.com, howtogeek.com, huffpost.com, hulu.com, imdb.com, inc.com, instacart.com, investopedia.com, lemonde.fr, lifehacker.com, makeuseof.com, marketwatch.com, mashable.com, nature.com, nbcnews.com, netflix.com, newsweek.com, newyorker.com, nytimes.com, pcmag.com, pinterest.com, reddit.com, reuters.com, sciencedirect.com, smh.com.au, spiegel.de, techcrunch.com, telegraph.co.uk, theage.com.au, theatlantic.com, theregister.com, thetimes.co.uk, theverge.com, travelandleisure.com, tripadvisor.com, and 7 more in the full dataset.
Sites that allow cohere-ai by name
airbnb.com, amplitude.com, cloudflare.com, deel.com, delta.com, jasper.ai, render.com, soundcloud.com, squarespace.com, stability.ai.
See all 30 AI crawlers, the rule change log or the full study. Being reachable is only the first step: how AI search engines choose what to cite covers the rest.
Cite this data
You are welcome to quote or republish these figures. Please credit IndexHalo with a link to this page (CC BY 4.0). All our datasets are on the data page.
IndexHalo. "cohere-ai block rate across major websites." Measured 3 November 2026. https://www.indexhalo.com/ai-crawlers/cohere-ai/
<a href="https://www.indexhalo.com/ai-crawlers/cohere-ai/">cohere-ai block rate across major websites, IndexHalo</a>