Industries / Finance and business services / Advertising and marketing / Graphic and web design
Brand identity
Brand identity is the work of designing a client's logo, colors and typefaces, and the rulebook, called brand guidelines or a graphics standards manual, that keeps them the same wherever they appear.
6
Milestones achieved, from 1876 to 2025.
3
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1876
UK trade mark number one
When the Trade Marks Registration Act 1875 came into force, a British business could register its mark with the state, and on 1 January 1876 the Bass & Co red triangle label became the UK's first registered trade mark. A registered mark is property its owner can defend. The UK Intellectual Property Office says the register now protects more than 2.5 million marks, received around 200,000 applications in the past year, and still holds more than 400 marks filed before 1900.
Credit
Bass & Co; the UK Parliament, for the 1875 Act
1907
A company planned as one
The Berlin electrical company AEG made the architect Peter Behrens its artistic consultant, and from 1907 to 1914 he designed its products, advertising and buildings, including the AEG Turbine Factory of 1909, by the Vitra Design Museum's account. A history page on an AEG website calls him "the inventor of corporate identity," meaning a company's whole appearance planned as a single system, and says a large company's image was "comprehensively defined and standardized for the first time." Those two claims are the company's own.
Credit
Peter Behrens; AEG
1956
IBM's first design program
Thomas J. Watson Jr., who had taken over IBM from his father, set up its first Design Program and hired the graphic designer Paul Rand, who redrew the IBM logotype (a logo made only of lettering) in a typeface called City Medium. In 1972 Rand drew it in eight horizontal stripes, a version IBM says has stayed "unchanged since 1972." The Henry Ford museum dates the two logotypes 1956 to 1972 and 1972 to the present.
Credit
Thomas J. Watson Jr.; Paul Rand; IBM
1975
NASA's worm and manual
The New York firm Danne & Blackburn designed NASA's red lettering logotype, nicknamed the "worm," introduced in 1975 and set out in a Graphics Standards Manual of 1976. In 1992 NASA retired it, except on clothing and souvenirs, in favor of its original round insignia from the late 1950s, the "meatball." It came back in April 2020 for the first crewed SpaceX launch, Demo-2, and NASA's brand guidelines now allow it as a supplemental graphic.
LessonA rulebook protects a mark only as long as the client's leaders follow it; NASA's administrator Jim Bridenstine said in 2020, "The NASA style guide is very clear, the worm no longer exists — but I write the style guide.".
Credit
Richard Danne and Bruce Blackburn; NASA
2010
Gap's logo lasts a week
In October 2010 the clothing chain Gap replaced its decades-old logo, white letters on a navy blue box, with black letters on white and a small blue box on its website, without announcing it. Customers complained on Twitter, Facebook and blogs, and Gap first invited the public to submit designs. On 12 October, one week after the swap, it said it would keep the old logo, the Associated Press reported.
LessonHansen said, "There may be a time to evolve our logo, but if and when that time comes, we'll handle it in a different way," and that this was not the right project to offer up to "crowd sourcing.".
Credit
Gap Inc.; Marka Hansen, then president of Gap North America
2025
Cracker Barrel restores Old Timer
On Tuesday 19 August 2025 the US restaurant chain Cracker Barrel showed a new logo with just its name on a gold background, dropping the barrel, the words "Old Country Store" and the seated man known as Uncle Herschel. The Associated Press reported that its shares fell more than 7% by the close of trading that Thursday. On 26 August the company said: "Our new logo is going away and our 'Old Timer' will remain."
LessonFifteen years after Gap, a rebrand that removed a figure customers knew was still reversed within a week.
Credit
Cracker Barrel Old Country Store
Watch for
USPTO image search for marks
In April 2026 the US Patent and Trademark Office (USPTO) added a beta feature to its free trademark search: a user uploads an image and it returns registered marks with similar design elements. The law firm Akerman reports that the office "has not published detailed technical specifications for the tool's matching algorithm," and advises treating results "as a supplemental resource, not a substitute for a comprehensive clearance search."
Achieved whenan outside party tests it on decided cases and publishes how often it finds the earlier, conflicting mark, or the USPTO publishes accuracy figures that an outside test confirms.
Claimed by
the USPTO, April 2026, as reported by Akerman on 1 May 2026
AI reviewers of brand rules
Frontify, which sells software for keeping brand guidelines, says its Brand Check app "analyzes your assets against your guidelines and reports issues automatically," and describes an AI agent called Specs, "still in development," meant to review logo use and placement and check fonts after a rebrand. Adobe introduced Brand Intelligence on 20 April 2026, saying it "helps ensure every piece of content is on brand" by learning from reviewers' feedback, rejections and approvals. Neither company gives accuracy figures.
Achieved whenan outside party publishes how many known rule breaks a tool catches, and how many false alarms it raises, on a test set the vendor did not choose.
Claimed by
Frontify, page updated February 2026; Adobe, April 2026
A brand file for machines
The Ad Context Protocol (AdCP), a standard for advertising software agents, defines brand.json, a file a brand hosts on its own website that its documentation calls "the canonical source of brand identity data." It has fields for logos, colors, fonts and tone of voice, and for rules such as the minimum clear space around a logo (the empty margin it must keep) and which colors may be paired. The documentation also says: "Most brands return nothing — which means agents are guessing."
Achieved whenbrands outside the protocol's group publish the file on their own sites, and design or brand software made outside the group reads it.
Claimed by
AdCP and AgenticAdvertising.org, documentation for schema version 3.2.8, November 2026
Open
01
A scorecard for logo search
A public test of the free trademark image searches, the USPTO's beta and the World Intellectual Property Organization's Global Brand Database, on US cases with known outcomes. The cases are decisions of the Trademark Trial and Appeal Board (TTAB, the USPTO's tribunal) from 2025 and 2026 in which the owner of an earlier mark tried to block a new one, called an opposition, and both marks include a design. For each case, upload the newer logo and record whether, and at what rank, the earlier mark appears. Publish case numbers, serial numbers, query dates and ranks, not the logo images.
The briefCredit · unclaimed02
Planted errors for brand checkers
A made-up brand with a rulebook of about ten rules (palette colors, approved typefaces, minimum logo size, clear space, no stretching or recoloring the logo, minimum color contrast), written once as a PDF for people and once as data files: a design-tokens file plus brand.json-style logo rules. A script generates 500 layouts (posters, social media tiles, letterheads), each with no rule break or exactly one, logged as the answer key. Then score AI brand checkers and general image-reading AI models on it.
The briefCredit · unclaimed03
Brand books into data files
A test of tools that read a brand guidelines PDF and write out its rules as data. Take 30 guidelines their owners publish openly (public bodies such as NASA publish theirs) and type an answer key for each by hand: palette color codes, typeface names, minimum logo size and clear-space rule, in Design Tokens 2025.10 and brand.json fields. Score each tool, counting values left out and values made up separately. Publish the answer keys with links to the PDFs, not copies of them.
The briefCredit · unclaimed
Open 01 · the brief
A scorecard for logo search
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of the free trademark image searches, the USPTO's beta and the World Intellectual Property Organization's Global Brand Database, on US cases with known outcomes. The cases are decisions of the Trademark Trial and Appeal Board (TTAB, the USPTO's tribunal) from 2025 and 2026 in which the owner of an earlier mark tried to block a new one, called an opposition, and both marks include a design. For each case, upload the newer logo and record whether, and at what rank, the earlier mark appears. Publish case numbers, serial numbers, query dates and ranks, not the logo images.
Why it is still open
The USPTO tool is a beta whose matching method is unpublished. The one large test of trademark offices' search tools against legal decisions, by Vandamme, Cabay and Debeir at ICAIL 2023, used 8,126 EU opposition decisions to test the EU (EUIPO) and Benelux (BOIP) offices' tools; its authors suspected the EUIPO tool had been trained on those same decisions and called for "another set of decisions." The same paper notes that the main US study covered word marks only. A 2017 dataset built from TTAB cases predates the current AI tools and tested older image-matching methods.
Built with what exists
USPTO trademark search (free account); WIPO's Global Brand Database, free, with image search since 2019; public TTAB decisions; a spreadsheet or a short Python script for scoring. Queries run by hand or within each site's terms of use.
First step
Pick 30 TTAB final decisions from 2025 and 2026 that involve design marks, run them through the USPTO beta by hand, and publish the pilot sheet.
Done looks like
At least 200 decided cases; for each tool, the share of cases with the earlier mark in the top 10 and top 50 results and its mean rank, split by whether the board found confusion, the same measures as the 2023 EU study; a second person re-checks 10% of the queries; a repeat run three months later.
Open 02 · the brief
Planted errors for brand checkers
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A made-up brand with a rulebook of about ten rules (palette colors, approved typefaces, minimum logo size, clear space, no stretching or recoloring the logo, minimum color contrast), written once as a PDF for people and once as data files: a design-tokens file plus brand.json-style logo rules. A script generates 500 layouts (posters, social media tiles, letterheads), each with no rule break or exactly one, logged as the answer key. Then score AI brand checkers and general image-reading AI models on it.
Why it is still open
The authors of Graphic-Design-Bench, a 49-task benchmark for AI models on design work published in March 2026, write that for "checking color compliance against brand guidelines, these failure rates are not acceptable margins of error." The BrandGuide dataset of June 2026 lists automated checking of design assets against guidelines only as a future use. Frontify's reviewer is still in development and gives no accuracy figures.
Built with what exists
Penpot, a free, open-source design tool with design tokens built in, or SVG files (a text-based vector image format) written by a Python script; the Design Tokens 2025.10 format; the brand.json schema for logo rules; vendor trials where available, such as Frontify Brand Check; Gemini Decision 1, which returns a probability for each fixed yes-or-no question such as "Is the logo below minimum size?" (see the Note "A thousand questions, one pass").
First step
Write the made-up brand's ten rules as a tokens file and a brand.json file, generate 50 layouts, and run one checker on them.
Done looks like
500 layouts, at least 40 per rule and at least 100 with no break; each checker's catch rate and false-alarm rate per rule, for at least three checkers; two human designers label a 100-layout subset so the tools can be compared with how often people agree; all files under an open license.
Open 03 · the brief
Brand books into data files
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A test of tools that read a brand guidelines PDF and write out its rules as data. Take 30 guidelines their owners publish openly (public bodies such as NASA publish theirs) and type an answer key for each by hand: palette color codes, typeface names, minimum logo size and clear-space rule, in Design Tokens 2025.10 and brand.json fields. Score each tool, counting values left out and values made up separately. Publish the answer keys with links to the PDFs, not copies of them.
Why it is still open
The one open-source converter found, pdf-brand-to-json, warns users to treat its output "only as suggestions" and has no test set. ExtractBench, a February 2026 benchmark for turning PDFs into structured data, uses 35 PDFs from other fields, such as financial reports. The brand.json documentation says most brands publish no machine-readable file at all.
Built with what exists
the Design Tokens 2025.10 format; the AdCP brand.json schema; ExtractBench's scoring method; pdf-brand-to-json; an open-weights model such as Mistral Large 4, so others can rerun the same model (see the Note "A trillion open parameters").
First step
Collect 10 openly posted guidelines and type the answer keys for two of them.
Done looks like
30 guidelines and at least 600 hand-checked fields; accuracy per field, with the rates of left-out and made-up values, for at least three tools.