# Editorial standards for SaaS content teams > What belongs in a SaaS style guide: claim standards, statistic sourcing rules, screenshot specs, a fact check protocol and a scored editing rubric writers can use. Source: https://saas-marketing.net/guides/saas-editorial-standards/ Topic: SaaS Content Marketing Type: guide Published: 2026-09-11 Last updated: 2026-09-11 Publisher: SaaS Marketing (saas-marketing.net) License: CC BY 4.0. Quote or republish with attribution and a link to https://saas-marketing.net/guides/saas-editorial-standards/ ## Short answer A SaaS editorial style guide is a written set of rules that a freelancer can follow without asking questions. It covers voice and sentence-level mechanics, a claim standard requiring every statistic to carry a primary source with a date and a link, product accuracy and screenshot specifications including version dating, competitor claim rules with a legal review trigger, and a scored editing rubric with a stated pass threshold. It exists to make quality repeatable when volume rises. ## Key takeaways - An unsourced statistic is a liability now, because answer engines quote the sentence and attach your brand name to it. - Secondhand blog citations fail the claim standard: trace every number to the organisation that ran the study. - Screenshots need a version number and a capture date in the alt text, or they rot inside six months. - Score drafts out of 40 across accuracy, structure, specificity and usefulness, and reject anything under 30. - Any competitor claim about pricing, security or performance triggers legal review before publication, no exceptions. - A one page cheat sheet gets followed. A 40 page brand bible gets skimmed once and then ignored. --- Most SaaS style guides are 40 pages of brand adjectives and nobody reads past page three. The useful version is closer to two pages of rules plus a rubric, and it exists for one reason: a freelancer who has never used your product should be able to file something publishable on the first attempt. That is the test. If a new writer needs a call to understand what you want, the guide has failed and you are paying for that gap on every single brief. ## Why editorial standards changed in 2026 The cost of a sloppy claim went up. When Google's AI Overviews or Perplexity quote a sentence from your page, they lift the sentence and attach your domain to it. Your brand is now the named source for a number you copied from an agency blog that copied it from somewhere else. BrightEdge measured AI Overviews appearing on roughly 82% of B2B technology queries in 2025, and Ahrefs put the organic click-through drop on affected queries at about 61%. Fewer clicks, more quotation. That flips the incentive: the page's job shifts toward being quotable and correct rather than being clicked. So the claim standard stops being editorial hygiene and becomes risk management. An unsourced statistic used to be filler. Now it's a liability with your logo on it. If a model quotes your page, it quotes one sentence without the surrounding qualification. Write every factual sentence so it survives being extracted alone. ## What goes in the voice and mechanics section Keep this part short and testable. Rules a writer can check against their own draft in ninety seconds, not adjectives. Here is the version we would hand a freelancer working on [SaaS content marketing](/saas-content-marketing/) pages: - Second person, present tense, contractions on. "You'll need two engineers" beats "organisations will require engineering resource." - No sentence over 35 words. Vary length deliberately: put a short one after a long one. - Define an acronym once, on first use, then use it freely. - Numerals for anything countable. Spell out only at the start of a sentence. - British or American spelling, pick one, state it, and note it in the brief so nobody guesses. - No sentence may open with a participial clause more than twice per piece. Then add the banned list. Every team has its own, and yours should be built from your actual rejected drafts rather than copied from a blog post. Ours runs about 60 entries and the top offenders are the phrases that signal a writer is padding: generic scene-setting and empty transitions. If a phrase appears in three rejected drafts, it goes on the list. One more rule that saves more time than any other. Every section must contain something that could not appear in an article about a different topic: a number, a named product, a price, a real objection. Writers self-edit against this and paragraph count drops by a third. ## The claim standard: what counts as a source A statistic is publishable when it carries four things: the organisation that collected the data, the year, the sample size or method, and a live link to the primary document. Miss any one and the claim is rejected rather than queried. | Claim type | What the writer must supply | Editor action | | --- | --- | --- | | Published research statistic | Organisation, study name, year, sample size, direct link | Open the link, confirm the number and the year | | Vendor benchmark report | Report title, year, methodology note, link | Flag vendor incentive in the sentence itself | | Pricing figure | Vendor pricing page URL, date checked | Recheck if older than 90 days | | Practitioner range | Stated as a range, labelled as an estimate | Accept with the estimate label attached | | Arithmetic done on the page | Inputs and formula shown to the reader | Recompute once | Secondhand citations fail. A blog citing a blog is not a source, and the SaaS category is full of numbers that have been laundered through five hops until nobody can find the origin. The famous organic-versus-paid CAC figures in circulation are a good example: widely repeated, rarely traced, and almost never segmented by ACV band. If you cannot get to the original, either drop the number or restate it as a practitioner range and label it honestly. Two allowances keep this workable. Practitioner ranges are fine when you label them as ranges built from aggregated reports rather than a study. Arithmetic is fine when you show the inputs on the page so a reader can rerun it with their own numbers. What is never fine is inventing a figure and attaching it to a named organisation, which is the single failure mode that ends a reference site's credibility permanently. If you are tracking your own numbers, publish them properly through something like [SaaS content marketing benchmarks](/research/saas-content-benchmarks/) rather than dropping them into prose unsourced. ## Product accuracy and screenshot specifications Screenshots rot faster than prose. Pricing pages, onboarding flows and settings screens change every quarter, and a stale screenshot on a high-traffic page costs you more trust than a typo ever will. The spec we would enforce: **Screenshot requirements** Run a quarterly sweep of your top twenty pages by traffic. Anything captured more than two product releases ago goes into a reshoot queue. Teams that skip this discover the problem through a support ticket, which is the expensive way. Product naming needs its own short list. If the product calls it a Workspace, the content calls it a Workspace, capital W, every time. Ahrefs is consistent about this, Notion is consistent about this, and the reason both read as credible is partly that their docs and their blog use identical nouns. Inconsistent product nouns are the fastest way to signal that marketing does not talk to product. ## Competitor claims and the legal review trigger Comparison pages convert better than almost anything else you publish, and they carry the only real legal exposure in your content programme. The rule is simple: factual, current, sourced, and dated. A claim about a competitor goes to legal review when it touches any of these: - Pricing, including a claim that you are cheaper - Security posture, certifications or compliance status - Performance benchmarks of any kind - Uptime, reliability or support response times - Anything describing the competitor's business as failing, shrinking or unstable Everything else, meaning feature presence, documented limits and publicly stated positioning, ships on the writer's sourcing alone. Cite the competitor's own documentation and date the citation in the page. "As of March 2026, Semrush lists X on its pricing page" is defensible. "Semrush is more expensive" is not, because it will stop being true and you will not notice. Set a review cadence of 90 days on every comparison page and put the owner's name on it. The pages that make you money are the pages that go stale fastest. A competitor changes pricing, your comparison page keeps the old number, and their legal team sends a takedown letter. The page comes down for two weeks during your best quarter. Dating every claim is what prevents this. ## The scored editing rubric Give editors a number, not a feeling. Four dimensions, ten points each, 30 out of 40 to publish, and nothing below six on any single dimension. Accuracy below six is an automatic rejection regardless of the total. A well-structured piece with a wrong number does more damage than a rough piece with right ones, because the well-structured one gets quoted. Editors should record the score in the CMS. After three months you will know which writers score consistently above 32 and can move to spot-check-only review, which is where the real throughput gain in any [content engine that ships twelve pieces a month](/playbooks/saas-content-engine/) comes from. It is not faster editing, it is fewer pieces needing deep editing. **30/40** Minimum publishable rubric score, with no dimension below 6 ## The one page cheat sheet Writers will not read the full guide before every piece. They will read one page if it fits on one page. Ours has five blocks: **Cheat sheet structure** Attach it to every brief. If you use an [editorial calendar template for SaaS teams](/templates/saas-editorial-calendar-template/), put the cheat sheet link in a column so it travels with the assignment rather than living in a folder nobody opens. ## The rejection checklist Editors need permission to send work back without writing an essay. Give them a numbered list and let them reply with numbers. **Reject the draft if any of these are true** Ten items, ten seconds to check, and the writer gets specific feedback without the editor drafting a paragraph. This matters most with freelancers, where feedback quality determines whether draft two is better or merely different. ## Where AI drafted work fits Generated drafts fail this standard in a predictable place. Prose quality is fine. Citation integrity is not. Models produce confident, well-formed statistics attributed to real organisations that never published them, and an editor reading for flow will sail straight past. So the rule for [AI assisted content production for SaaS](/guides/ai-assisted-content-production/) is narrow: generation is allowed for structure, first drafts and variations, and every number in the output is treated as unsourced until a human opens the primary document. No exceptions for numbers that look right. The plausible ones are the dangerous ones. If you are deciding where machine drafting belongs in your workflow at all, the tradeoffs are laid out in [AI written versus human written content](/comparisons/ai-written-vs-human-written-content/). The short version: generation moves the bottleneck from writing to verification, and your rubric is what stops that bottleneck from being ignored. The same discipline applies to customer-sourced material. A quote pulled from a call needs approval and a role title, and the process for that belongs in your [customer story engine](/playbooks/customer-story-engine/) rather than in the writer's head. ## What this costs and what it does not fix Writing the guide takes a senior editor about two days. Enforcing it costs roughly ten minutes of spot checking per draft once writers supply source lines, and considerably more in the first month while people adjust. What it does not fix: a weak topic. A perfectly sourced, well-structured article about a subject nobody searches for and no buyer cares about is still a waste of the budget. Standards raise the floor on execution. They do nothing for selection, which is a strategy problem covered in [defining your audience, problem set and point of view](/courses/b2b-saas-content-strategy/01-define-your-strategy/). The other honest cost: some good writers hate rubrics. You will lose one or two who feel scored rather than trusted. That tradeoff is worth taking when you publish more than six pieces a month, and probably is not below that, where an editor can hold the standard in their head. ## Start here Write the one page cheat sheet first, before the full guide. Use it on the next five briefs, note every question a writer asks anyway, and those questions become the sections of the full document. Then set up the rubric in whatever your team already uses so scores land next to the draft rather than in a separate sheet, which is where [the SaaS content marketing tool stack](/tools/saas-content-marketing-tools/) matters more than the specific vendor. Give it a quarter. If editing time per piece has not dropped by a third, the guide is too long and writers are ignoring it. ## Frequently asked questions ### What should a SaaS content style guide actually contain? Five parts: voice and mechanics rules a writer can apply sentence by sentence, a claim and sourcing standard, product accuracy rules covering screenshots and UI naming, competitor claim rules with a legal escalation trigger, and a scored editing rubric. Anything beyond those five parts belongs in a separate brand book that writers never need to open. ### How do you fact check content at scale? Require the writer to submit a source line beside every number: the organisation, the study name, the publication year, the sample size and a direct link. The editor spot checks three claims per draft against the primary source. If any one fails, the whole draft goes back. That rule keeps checking cost near ten minutes per piece instead of an hour. ### Are secondhand citations acceptable in B2B content? No. A blog citing a blog citing a vendor report is how wrong numbers survive for a decade. Trace the claim to the organisation that collected the data, confirm the year and the sample, and link to that source. If the original is paywalled or has vanished, drop the claim or restate it as a practitioner range with the estimate labelled as such. ### How do you keep product screenshots accurate? Store the product version and capture date with every image, name the owning team, and run a quarterly sweep of the twenty highest-traffic pages. Any screenshot older than two product releases gets reshot or removed. Most teams find their worst offenders on pricing and onboarding pages, which change most often and rank best. ### What is a reasonable pass threshold for an editing rubric? Score four dimensions out of ten each: factual accuracy, structure, specificity and usefulness. A draft needs 30 of 40 to publish, with no single dimension below six. Accuracy below six is an automatic rejection regardless of the total, because a well-structured piece with a wrong number is worse than a rough piece with right ones. ### Should AI drafted content follow the same editorial standard? The same standard, applied harder. Generated drafts invent plausible statistics and attribute them to real organisations, which is precisely the failure your claim standard exists to catch. Treat every number in an AI draft as unsourced until a human opens the primary source. Most teams that publish generated content at volume fail on citation integrity, not on prose quality. ### How often should a style guide be updated? Twice a year for the rules, monthly for the examples. Rules change slowly. The examples of good and bad writing go stale fast, and writers learn far more from three before-and-after pairs pulled from last month's drafts than from another paragraph of abstract guidance.