Guide · updated October 2026

How to turn a brand guide PDF into email QA rules

The copy deck is the first contract; the brand guide is the second. It runs forty pages, it was written for print and web, and everyone read it once, the week they were onboarded. Email is where it gets violated most, because templates get cloned, fonts fall back, and logos get swapped by whoever built the last campaign. The fix is not re-reading the PDF before every send. It is extracting the rules once, as values a machine can test, and enforcing them on every build.

This guide covers what a brand guide actually says about email, the rulebook field by field with what each rule catches, the two ways to get the rules in (the PDF, or an approved email), why provenance matters, and what fails most in real campaigns.

What a brand guide actually says about email

Most of the document is irrelevant to an email build: print bleeds, signage, the tone of voice essay. The parts that matter are small and precise, and they are the parts that drift:

• Palette. Primary, secondary and background colours as exact hex values. An email inherits a “close enough” blue from a cloned module and nobody sees it until the client does.
• Typography. Families, the weights that are actually licensed, sizes, and the web-safe fallback stack. A headline that renders in a fallback weight passes every visual read-through and still is not the brand.
• Logos. Which variant goes on which background, minimum size, and dual-logo lockups for co-branded sends. The header and footer are where the wrong file hides.
• Trademarks and business names. Where ® and ™ appear, that they render as true superscripts rather than inline text, and how the product and company names are spelled and capitalised.
• Layout and spacing. Template width, module padding, button dimensions: the numbers that make two campaigns look like the same brand.
• Links. The domains the brand owns, the ones that must never ship (staging, test, last year’s microsite), and whether each language links to its own pages.
• Copy conventions. Capitalisation rules and approved button wording. “Learn More” versus “Learn more” is a brand decision, not a typo.

Why the PDF can’t be the rulebook

A rule only protects you when it is an explicit value something can test against. “Use the primary palette” is guidance; #C41F3E is a rule. “Body copy in the brand face” is guidance; the brand face at 400 with an Arial fallback, 16px, 1.5 line height is a rule. Extraction is the act of turning the first kind into the second, and doing it once, so every campaign is checked against the same numbers instead of each reviewer’s memory of the document.

The rulebook, field by field

This is the shape of a complete email rulebook for one client, with what each group of rules catches once it runs as a check. Most clients fill in half of it and get most of the value.

Two ways to get the rules in

From the brand guide PDF. Colour swatches and dimension callouts are graphics, so the text layer of the PDF cannot see them at all; the pages have to be looked at. SendLint rasterises the guide and the AI Agent reads it six pages at a time, up to thirty pages (real guides run 20 to 40 MB and that is fine). Every value it proposes carries three things: the page it was read from, the quoted source text, and a confidence level. Values it is unsure about land in a separate “found but unsure” group so they get looked at first.

From an approved email. When there is no guide, or the guide predates the email templates, start from a build the client has already signed off. SendLint reads its style fingerprint, the colours, fonts, sizes and widths the build actually uses, with how often each was seen and where (“seen 14 times, .cta a”), and proposes the same rulebook with that as the evidence. Precedence is fixed: an authored guide outranks a sample, and a sample outranks SendLint’s own defaults, so a sample is a starting point, not a ceiling.

Both paths end in the same review list. Nothing is saved until the whole list has been reviewed and confirmed. A value you edit by hand leaves the review list, because you have now taken responsibility for it.

Provenance, or it didn’t happen

The failure mode of automated extraction is confidence without evidence: a palette of six hex values and no way to know which page they came from, or whether one is the designer’s example rather than the spec. Page, quote and confidence on every value is what makes the review a two-minute job instead of a second reading of the guide, and it is what lets you answer the client when they ask why the build was flagged: page 14, “Legal copy is set in Brand Grey 60%”.

From rules to checks

Once the rules exist, they stop being a document and start being checks that run on every email: colour compliance against the palette and the role colours, font family and weight verification, the email width, dual-logo separation, superscript placement and weight, trademark fidelity, CTA case, business-name capitalisation, bold punctuation, the client’s grammar rules, link domains and language. Two properties matter here:

• A check with no rule says so. Colour compliance with an empty palette would silently pass everything. It shows as “Not configured” and points at the exact section to fill in, and the report says in one line that brand checks were skipped.
• Rules are per client. An agency runs a separate rulebook for each brand, copies a configuration to bootstrap a new client, and sets each check to On, Warn or Off per client, because one brand’s hard rule is another’s preference.

What fails most in real campaigns

• Legal copy in a grey that is not the brand’s grey, and under the contrast floor as well. The accessibility checklist catches the second half.
• Headlines in the fallback weight because the licensed weight was never loaded in the template.
• A bold, inline ® that breaks the line it sits on, or sits in the hero headline where the guide forbids it.
• “Learn More” on every button, title case, in a brand that writes buttons in sentence case.
• The partner logo touching the client’s in a co-branded header.
• A French CTA landing on the English page, and a link to the staging domain that opens fine inside the agency.
• A product name without its mark in one module and with it in the next.

Done by hand vs done by machine

This is what SendLint’s Brand rules do. Upload the client’s brand guide PDF, or paste an approved email, review the proposal, confirm, and from then on those rules run as checks on every email for that client, next to the copy deck comparison and the rest of the email QA checklist. Your brand review becomes a review of findings instead of a re-read of the guide.