How it works

From your guide to a checked file.

Publish your reviewed brand rules once, then use them while creating work in ChatGPT or Claude.

Brand guide · PDF
Template · .potx

  1. For you: Compile

    Keldane Teal#0E7C7Bp.14 §2.03

    Each value cites its page or is flagged.

  2. For your people: Serve

    • ChatGPT
    • Claude
    • API

    Your admin connects ChatGPT and Claude once.

  3. For your people: Build

    Built on your template.

  4. For your people: Check

    • Georgia → Keldane Display
    • #14A3A1 → Keldane Teal

    Fixed, or sent back.

  5. For anyone they send it to: Certify

    BW-3C1D-08AA

    Signed. Anyone can verify it.

1 Compile

Upload your brand guide (PDF) and your PowerPoint templates. Each value Brandwarden keeps cites the page or file it came from, or is flagged as an assumption.

When sources disagree, nothing publishes until you decide. You can save a question for later, and it keeps blocking, or record an exception with an end date. There are four kinds of question: two values that disagree, a name the palette doesn't define, a font weight that doesn't exist, and a size with no axis or unit.

A reading uses the sources you confirm. Approved assets can be supplied to a connected assistant; font files require separate owner approval for distribution.

Keldane Teal

#0E7C7B

p.14 §2.03

brand

Pantone · not in guide

Details

RGB 14 124 123
CMYK 89 0 1 51

Harbor Navy

#14213D

p.14 §2.04

brand

Pantone · not in guide

Details

RGB 20 33 61
CMYK 67 46 0 76

Sandbar

#E9C46A

p.41

charting

Pantone · not in guide

Details

RGB 233 196 106
CMYK 0 16 55 9

Signal Red

#C8102E

p.41

charting

negative values only

Pantone · not in guide

Details

RGB 200 16 46
CMYK 0 92 77 22

More detail: type styles and a review question

Title

Quarterly partnership review

Keldane DisplayBold · 28/32 pt

Named on p.22

Body

Keldane Marine serves 1,400 ports in 62 countries.

Keldane SansRegular · 11 pt

p.22

Caption

Source: Keldane annual report, 2025

Keldane Sans9 pt · +0.5 pt tracking

Assumed

1 of 1 decisions made

Published · version 1.0

Chart colors: the guide's order and the template's order differ

Guide p.41 · keldane.potx · value mismatch

ResolvedSaved for later

2 Serve

When someone asks ChatGPT or Claude for a deck or a document, the assistant is told to load your rules first: a short brief1 with exact values, the layouts it may use and your voice rules. Your own apps get the same brief through the API.

Ask the assistant to use your published Brandwarden rules and approved assets. The host creates the content; a Brandwarden build is optional.

¹ About 2,500 tokens, small enough to travel with every request.

brief · deck · Keldane · standard

Follow these rules exactly; the file is checked against every one of them before it is certified.

- Title / section header: Keldane Display Bold, 28/32pt, Keldane Teal. MUST be ≤ 60 characters.

- Series colors MUST follow the sequence: 1 Keldane Teal #0E7C7B → 2 Harbor Navy #14213D → 3 Sandbar #E9C46A.

- MUST use the active voice. MUST define every acronym on first use.

3 Create

ChatGPT or Claude creates the content using your published brand rules and the approved assets available to that connection. The host can use its own document, presentation or image tools; a Brandwarden build is not required.

Templates and logos carry their source and intended use. Fonts are shared only after separate owner approval of distribution rights. If an asset cannot be fetched, the assistant must say so rather than substitute a guessed brand asset.

keldane.potx

  • Title
  • Section
  • Two content
  • Chart

What the assistant sends

Layout: Chart

Title: Partners by region

Callout: Every partner met its SLA this quarter.

4 Check

A file Brandwarden builds is checked against your published rules. Fonts, colors and other mechanical problems are fixed automatically where possible. Your policy decides what else holds the certificate.

PowerPoint files made elsewhere can be checked too: the assistant runs a small script on the file and sends Brandwarden a summary. You get a report of the findings, never a certificate, because Brandwarden never saw the file.

Policy level

Fixes the mechanics and certifies what passes. Voice findings are notes, but if your guide warns against hype, hype words still hold the file.Fixes the mechanics. Superlatives and undefined acronyms must be fixed before certification.Every voice finding must be fixed, including passive voice. Fixes are still applied and reported.

On this build

fontsGeorgia in the title → Keldane Displayfixedfixedfixed

colorsTitle in #14A3A1 → Keldane Tealfixedfixedfixed

voiceSlide 3: “SLA” is not defined in the deck or the glossary.notedsent backsent back

voiceSlide 4: 2 of 3 sentences are passive.notednotedsent back

Certificate

Certified2 fixed · 2 notes
Held1 rewrite sent back
Held2 rewrites sent back

Fonts and colors are fixed at every level. At every level, an error we can't fix holds the file, and so do hype words if your guide warns against hype.

5 Certify

A file that passes gets a signed certificate inside it, in PowerPoint, Word or PDF. Anyone can check it on the verify page and gets one of four answers: certified, modified (with the changed slides named), invalid, or no certificate.

Only a file Brandwarden built can carry a certificate. A file made elsewhere gets a report of findings instead.

After one edit

Q3 Partnership Overview.pptx Modified

Changed after certification — slide 4

Certificate
BW-3C1D-08AA
Brand
Keldane
Brand version
1.0
Issued
Sep 12, 2026
Built in
ChatGPT
Policy
Standard
Changed
slide 4
  • Certified · exactly what was signed
  • Modified · changed afterward; the changed slides are named (PowerPoint)
  • Invalid · the signature doesn't verify
  • No certificate · nothing to check

Make Brandwarden available to your team

Brandwarden supplies the published authority; ChatGPT or Claude creates the content. Each person connects their own account and the available connection is shown in Brandwarden. A host-created file is not automatically certified. PowerPoint files made elsewhere can be checked through a summary produced by the extraction script; that path returns a report, not a certificate.

Rollout

  1. Create an account and workspace; add the brand sources you have.
  2. Review what was read, resolve questions and publish your brand rules.
  3. Use the connection instructions shown in Brandwarden for your ChatGPT or Claude account.
  4. Each person connects from their assistant with access to the workspace. Plan and seat rules follow the workspace's actual subscription; connecting an assistant does not silently change those terms.

Questions

Which tools does it work in?

ChatGPT and Claude workspaces, plus our API for your own apps. PowerPoint files made elsewhere can be checked.

What happens to our brand guide?

Brandwarden's own compiler reads it. Its page images go to an AI provider only when you check “Read with AI” for a compile, and every such reading is recorded. The guide itself stays in your organization's encrypted storage.

Does Brandwarden see our documents?

Content sent to Brandwarden is processed for the requested service. Built files and thumbnails are retained for download for 30 days by default. A request held before building because of capacity or plan limits can retain its full input for up to 30 days, until resumed, discarded or expired. Receipts and certificates remain as records. Content kept by ChatGPT or Claude follows that host's own policies.

What about our licensed fonts?

Font uploads are stored encrypted. Upload license attestation and approval to distribute a font file are separate: an owner must approve distribution rights before the font can be shared through the connector. The receiving host and viewer must support the font.

How do employees turn it on?

An admin shares the connection instructions. Each person connects from ChatGPT or Claude with their own Brandwarden access. Ask the assistant to use your published Brandwarden rules and approved assets. The host creates the content; a Brandwarden build is optional.

Can we verify a file someone sends us?

Yes. Drop a PowerPoint, Word or PDF file (up to 50 MB) on the verify page. You get certified, modified (with the changed slides named), invalid, or no certificate.

Sources

The two survey findings on the home page, with fieldwork and release dates. Neither tested Brandwarden.

  1. Templafy / Sapio Research, The Business-Ready AI Gap (fieldwork August 2026, published October 1, 2026, 2,000 respondents, US and UK)
    https://www.templafy.com/news/knowledge-workers-edit-ai-generated-documents/
  2. Optimizely / Savanta, 2026 global study (fieldwork May–June 2026, published June 30, 2026, 2,003 respondents, seven markets)
    https://www.optimizely.com/company/press/2026-global-data-study · https://www.optimizely.com/field-notes/guides/optimizely-2026-global-study

Ready when your guide is.

Start with your guide

Start with the access shown in your workspace. Available plans and any setup or trial allowances are shown before you agree to them.