Observed or maintained: where does brand context come from?

Six of the endpoints on this site read the public website of a company. This site calls that observed context. Two endpoints return context that comes from sources added for a brand, such as its website, guides, and files. This site calls that maintained context. The vendors do not use these terms. The kind of context decides what a response is good for.

Five questions to ask any response

1. Where does the context come from? It can come from one page, a whole site, or sources that someone added for the brand.
2. Which guidance or limits does it carry? It can carry a description of the brand, or rules that a model can follow.
3. Does it return files, links to files, or neither? If it returns links, do the links stay valid?
4. Does the endpoint document a version that you can refer to? Some give only a date, and some give nothing.
5. Can you ask for one task, or do you always get the whole profile?

This table gives the answers for the response of each endpoint for linear.app. The answers come from the responses and from the documentation of each vendor.

Endpoint	Comes from	Guidance	Files	Date or version	One task
Brandfetch Brand Context API	The public site	A voice summary and two tones to avoid	None	Read date	No
Brandfetch Brand API	The public site	None	Logos, symbols, and a banner on the Brandfetch CDN (content delivery network)	None for the data	No
Context.dev Brand	The public site	None	Logos and backdrops on media.brand.dev	Cache age	No
Context.dev Styleguide	The CSS of the site	None	Font files	Cache age	No
Firecrawl branding	One page	Three personality words	Inline logo, page images	None	No
Logo.dev Brand API	The public site	None	Logo, brandmark, and banner on links that expire	Index date	No
Rainbrand Brief	Sources added for the brand	Voice, look, and claim limits for the task	Assets selected for the task	Version used (skillId)	Yes
Rainbrand Brand Skill	Sources added for the brand	Eight guidance files	Files with signed links	Version ID	No

When is observed context enough?

Observed context is enough when you do not work with the company yet. Examples are prospect research, a personalized first line in an email, and a signup screen that you fill in from a work email. Another example is adding company data to a sales record. For these jobs, the website is the best source that you have. Brandfetch, Context.dev, Firecrawl, and Logo.dev return it for any domain. Of these, only the Brandfetch Brand Context API returns voice and positioning.

When do you need maintained context?

You need maintained context when work goes out under the name of a brand again and again. Examples are an AI product that makes ads for its customers, and an agency that runs client brands through Claude or ChatGPT. Another example is a team whose agents write for one company every day. For this work, the rules of the brand, its files, and a version that you can trace become important.

Some companies publish their rules. Linear does, on its brand page at linear.app/brand. None of the observed responses returned those rules. The observed responses also gave four different main colors for Linear. The maintained Brand Skill, added from linear.app, was the only response that showed which colors were official, as the Linear brand page does.

Can you use both?

Yes, and products often use both. They use an observed lookup at signup, when they have only a domain. They use a maintained brand after the customer adds their own sources. The customer brands recipe shows the change from one kind to the other in code.