Skip to content

Checks

E-E-A-T check

Checks a draft for a real author, backed experience from doing the work, depth an outsider couldn't fake, proof others would vouch for and claims that hold up.

Other tools

Add this skill to Claude

Claude takes skills as a ZIP file. Your download has started; if it didn't, use the button below.

  1. 1In Claude, open Customize, then Skills.
  2. 2Choose +, then Upload a skill, and pick eeat-check.zip.
  3. 3Ask for it by name, or let Claude use it when a task matches.

Skills need code execution turned on in Claude's settings. On a Team or Enterprise plan, an admin can share it with everyone.

Download the folder

Try it now, without installing

See the skill work on a made-up input before you add it. Nothing is installed, and you can swap in your own input.

Open in the Claude app

Opens a new chat in the Claude desktop app with the prompt filled in. Nothing is sent until you press enter.

Includes the whole SKILL.md and the sample input. Paste it into Claude, ChatGPT or any chat.

Influenced by
Search Quality Rater Guidelines, E-E-A-T, expertise, authoritativeness and trust
Needs
The draft or page · The author
Works with
Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md
Version
1.0.0 · Last checked · License MIT
Maintained by
Sabih AhmedContent
SkillContent

E-E-A-T check

Example output

Signals in one draft
PartStatusFinding
ExperiencePass2 backed signals
TrustNotes1 unattributed hedge
ReputationNot checkedNo browsing

Checks a draft for a real author, backed experience from doing the work, depth an outsider couldn't fake, proof others would vouch for and claims that hold up.

The skill

The whole skill, in plain text.

Nothing is hidden. Read it, change it to fit your team, or take the whole folder.

Scripts
Optional Python scripts for the counting. It works without them and says which way it ran.
Network
Opens web pages or your tools if your agent can. Works from pasted data if not.

3 test cases

eeat-check/SKILL.md

In the folder

  • eeat-check/SKILL.md
  • eeat-check/references/signals.md
  • eeat-check/scripts/eeat_signals.py
---name: eeat-checkdescription: >-  Checks a B2B SaaS article or page for the signals behind experience, expertise, authoritativeness and trust: a  real named author with a bio, first-hand experience that is backed rather than performed, depth a non-expert  couldn't fake, proof others would vouch for (independent reviews, coverage, the author's work elsewhere), accurate  sourced claims and honest limits, with stricter rules for money, legal, health and safety topics. Returns PASS,  PASS WITH NOTES or FAIL with a fix per finding, and questions for the author where only they can supply the proof.  Use it before an article ships or when a page needs to read as more credible. Not for checking tone (use  voice-check) or every figure's source in detail (use claim-check).license: MITcompatibility: >-  Works in Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md. Researches the company's and author's  reputation if the agent can browse; otherwise asks for links. Python 3 is optional, for counting signals.metadata:  author: SHAKEUP  version: "1.0.0"  stage: check  plugin: content  homepage: https://www.shakeuphq.com/skills/eeat-check--- # E-E-A-T check Finds where a draft claims credibility it hasn't shown, and where real credibility is missing, so the fix is proof, not adjectives. ## Purpose - One job: check one draft or page and return findings by part (experience, expertise, authority and reputation, trust), each with a fix.- Stage: check. It ends in PASS, PASS WITH NOTES or FAIL with severities.- It never writes experience, credentials, quotes, reviews or customers into the draft. Where proof is missing, it asks the author for it. ## Use when / not | Situation | Use || --- | --- || An article is ready for signoff | this skill, after `article-write` || A published page lost ground after a search update and the content may read as thin | this skill, with `refresh-audit` || "Make this more credible" or "review for E-E-A-T" | this skill || The wording doesn't sound like the company | `voice-check` || Every figure needs its source checked line by line | `claim-check` (this skill runs it or its test) || Structured data, author pages or site architecture | out of scope; note them as page-level items only | ## Inputs | Input | Required | Ways to provide, in order | If missing || --- | --- | --- | --- || The draft or page | yes | a file the agent can read; the live URL if browsing; pasted text | `BLOCKED`: ask for it || The author | yes | the byline; `## Content` > `Authors` in the profile; the reader's answer | `critical` finding: no named author || The author's bio and work elsewhere | no | the bio URL if browsing; links the reader pastes | reputation `not checked`; coverage `partial` || The company's independent reputation | no | search reviews and coverage if browsing; links the reader pastes | reputation `not checked`; coverage `partial` || The research brief | no | an `article-research` handoff | experience claims checked against the profile only || Approved claims | no | `## Claims and proof` | every first-party figure is unsourced until shown otherwise | ## Profile read From `company-profile.md`: `## Company`, `## Product`, `## Buyers`, `## Claims and proof` (with `### Never say`), `## Competitors` (mention policy), `## Content` (authors, topic clusters), and every `## Learnings` entry for `eeat-check` and `claim-check`. No profile: ask in one message for the author's name and bio link, links to independent reviews or coverage of the company, and the source of each figure in the draft. Say that `company-profile` saves this next time. Continue with coverage `partial`. ## Steps 1. **Apply learnings** for this skill (a finding the reader rejected before, an author's new credentials). 2. **Count what can be counted.** Run:    ```   python3 scripts/eeat_signals.py --draft draft.md --domain <site domain>   ```    Without Python, scan by reading for: first-hand phrasing ("we tested", "when we ran", "in practice"), observer phrasing ("many companies", "research shows", "experts agree"), "in our experience" followed by a generic statement, unattributed hedges, guarantees and superlatives, figures with no link in their sentence, money, legal, health and safety terms, the byline, a bio link and a date. Say which way you ran it. The counts are a first filter; the verdict comes from the steps below. 3. **Decide YMYL.** Use the terms from step 2 and the topic itself (`references/signals.md`, YMYL). When yes, apply the stricter rules to every later step. 4. **Who wrote it.** A real person, named, with a bio page; the expertise the draft implies matches what the bio and the profile say. A credential, title or experience in the draft that the bio or profile doesn't support is `critical`. A job title with no name is not an author. 5. **Experience.** List every experience claim. For each, find what backs it: the research brief, `## Claims and proof`, the author's confirmation. Backed: counts. Not backed: `critical` (invented). Then count the backed ones: fewer than 2 in an article is a `warning`, with a question for the author per missing spot ("which account showed this?"). 6. **Expertise.** Apply the practitioner test to each H2's first paragraph and each recommendation (`references/signals.md`). Flag sections with no condition, exception or mechanism, category descriptions with no position, and statistics with no reaction. 7. **Authority and reputation.** Check proof inside the page (cleared quotes, case studies, credible citations) and outside it (what independent sources say about the company and the author). Record each source found, with its link. None found outside the site is a `warning`; not checked is a coverage gap. Check the topic sits in an area the site already covers. 8. **Trust.** Run `claim-check` on the draft, or apply its test inline: every figure and comparison against `## Claims and proof` or a source opened in this run, every `### Never say` line, the competitor mention policy. Then check: guarantees, false precision, correlation as cause, dates, disclosure when the article recommends the company's own category or product, and whether each claim's basis (data, practice, research, opinion) is clear. 9. **Write the fixes.** For each finding: the line, the part, the severity, the problem in a few words, and the fix. When the fix needs proof only the author has, the fix is a question for the author, never a drafted line. ## Check and verdict Severity: `critical` (no named author; an experience, credential, quote or customer with nothing behind it; a figure with no source; a guarantee the page can't support; a `### Never say` match; for YMYL, any unsourced factual claim or no expert review note), `warning` (fewer than 2 backed first-hand signals; a section that fails the practitioner test; observer phrasing outnumbering first-hand phrasing; no bio link; no date; unattributed hedging; no reputation found outside the site; no disclosure of the company's interest), `suggestion` (a signal in a weak position, a stronger source available). | Verdict | Condition (first match wins) || --- | --- || `BLOCKED` | No draft || `FAIL` | 1 or more critical findings, or 5 or more warnings || `PASS WITH NOTES` | Warnings or suggestions only, fewer than 5 warnings || `PASS` | No findings | 5 warnings fail the draft because, together, they read as a page nobody stands behind, even when no single one is serious. No named stops. The calling skill (`article-write`, or the reader) applies the fixes and reruns. A request to add invented experience or a made-up quote is refused in the output, with the questions for the author instead. ## Learn Append to `## Learnings` when: - the reader rejects a finding as wrong for their market (a `proposal` with their reason);- the reader supplies proof: an author credential, a review page, a cleared case (a `fact`, and add it to the profile's `## Content` or `## Claims and proof`);- a page this skill passed later drops or gains after a search update (a `fact`, with the source). Example: `- 2026-10-24 | eeat-check | fact | Head of content's talk at a trade conference, linked on her bio | Why: reader supplied the link, 2026-10-24 | Status: noted` ## Output ```Verdict: <PASS | PASS WITH NOTES | FAIL | BLOCKED> · Coverage: <complete | partial (what wasn't checked)> · Signals counted: <script | by reading>YMYL: <yes | no> · Author: <name, bio link | missing>Counts: <n> critical, <n> warnings, <n> suggestions · First-hand signals: <n backed> of <n claimed> | Part | Status | Finding || Experience | <pass | notes | fail> | <one line> || Expertise | ... || Authority and reputation | ... || Trust | ... | | Line | Part | Severity | Problem | Fix |(critical first) Reputation found: <source, what it says, link> (or "not checked: <what would provide it>")Questions for the author: <one per missing proof>Learnings applied: <list, or none>``` Then the handoff block. ## Hand off ```handoffskill: eeat-checkversion: 1.0.0verdict: FAILcoverage: partial (reputation not checked: no browsing)for: <draft title>outputs:  critical: 2  warnings: 3  ymyl: false  author_questions: 2  fixes_ready: truenext: [article-write]open: [the author's example for the first H2]``` `article-write` reads the fix table in its loop; with no calling skill, `next` is `[]`. ## Failure modes | Problem | Do this || --- | --- || The byline is the company, not a person | `critical`: ask for the person who wrote or reviewed it || The author is real but writing outside their field | `warning`: suggest a reviewer with the right background, named in the byline as reviewer || A customer quote is real but not cleared | Treat as unbacked until the reader confirms clearance || The reader asks to add experience lines or an expert quote "to pass" | Refuse in one line; give the questions that would let the author supply them || Reviews exist but are poor | Record them as found; reputation is what others say, not only what helps || The draft is a product page, not an article | Check trust and the author or company only; experience and expertise checks apply to editorial copy | ## Influenced by - Google's Search Quality Rater Guidelines, which put trust at the center of E-E-A-T and send raters to independent sources for the reputation of a site and its content creators: https://guidelines.raterhub.com/searchqualityevaluatorguidelines.pdf- Lily Ray's work on E-E-A-T, including making the expertise of a brand's leaders and writers visible and being open about who runs the company: https://lilyray.nyc/e-a-t-expertise-authoritativeness-trustworthiness/- SHAKEUP's content engine: credibility shown through specifics rather than declared, the practitioner test, and stricter rules for money, legal, health and safety topics.

How it thinks

Inputs in, a verdict out, and a stop only where you decide.

What we built

Credibility is shown through specifics, not declared with adjectives. We added a firm line on proof. When experience is missing, the skill asks the author for it and never drafts a plausible line in its place.

It reads your company profile if you have one, or asks for only what it needs. Then it works through its steps, checks its own work and ends in a verdict.

The steps

  1. 01

    Count what can be counted

    Its script finds phrasing from doing the work ("we tested") and from reading about it ("research shows"), plus hedges, guarantees, figures with no link, the byline and a date. Counts are a first filter, never the verdict.

  2. 02

    Check who wrote it

    A real person, named, with a bio page. Team bylines and job titles with no name don't count as authors. Any credential the bio doesn't support is a critical finding.

  3. 03

    Test the experience

    Every claim of experience needs something behind it, in your profile, the brief or the author's own word. Unbacked experience is treated as invented.

  4. 04

    Test the depth

    Each section's opening is checked against one question. Could someone who has never done this work write it? If yes, it needs a condition, a threshold or a mechanism.

  5. 05

    Look for outside proof

    Reviews, coverage and the author's work elsewhere, with links. What others say counts for more than what the page says about itself.

  6. 06

    Check trust last and hardest

    Every figure against your approved claims or a source, plus guarantees, false precision, dates and a line saying so when the page recommends your own product.

What you need

  • Required

    The draft or page

    A file your agent can read → The live URL, if your agent can browse → Text you paste

  • Required

    The author

    The byline → The Authors line of your company profile → Your answer

  • Optional

    The author's bio and work elsewhere

    The bio page, if your agent can browse → Links you paste

  • Optional

    What others say about your company

    A search for reviews and coverage, if your agent can browse → Links you paste

What it can return

  • PASS

    No findings.

  • PASS WITH NOTES

    Warnings only, fewer than 5.

  • FAIL

    A critical finding, such as no named author, unbacked experience, an unsourced figure or a guarantee. Or 5 or more warnings.

  • BLOCKED

    There's no draft.

Where it stops for you

It runs through without stopping. You review the result.

Example output

What you get back.

A run for a made-up company. Your output depends on your inputs.

Here's a run for a made-up B2B SaaS company, on a blog post about attribution in Salesforce.

Verdict

PASS WITH NOTES

Coverage: partial (reputation not checked: no browsing)

PartStatusFinding
ExperiencePass2 backed signals: day-1 stage mapping and the setup figures, both in the approved claims
ExpertisePassNames when the advice fails: campaigns created manually, sales cycles under 30 days
Authority and reputationNot checkedNo links to reviews or the author's work elsewhere
TrustNotes"Self-reported answers might be less precise" has no source. Attribute it or commit to it
Questions for the author

which review site or talk should the bio link to?

The handoff block the next skill reads
skill: eeat-check
version: 1.0.0
verdict: PASS WITH NOTES
coverage: partial (reputation not checked)
for: How to report pipeline by channel in Salesforce
outputs:
  critical: 0
  warnings: 1
  ymyl: false
next: []
open: [links to the author's work elsewhere]

What good looks like

  • A named author and a bio link, or a critical finding saying they're missing
  • Every experience claim marked as backed or not, with its source
  • Outside proof listed with links, or marked as not checked
  • Questions for the author wherever only they can supply the proof

Install

Use it in the tool you already have.

Pick your tool. Each tab has the shortest path we know of.

  1. 01
    Download eeat-check.zip.
  2. 02
    In Claude, open Customize, then Skills.
  3. 03
    Choose +, then Upload a skill, and pick the ZIP. Code execution must be on in your settings.
  4. 04
    Ask for it by name, or let Claude use it when a task matches.

Questions

Before you add it.

Is E-E-A-T a ranking score?

No. It's how quality raters judge a page, and it describes what credible content looks like. This skill checks a draft against those signals so the page earns trust from readers first.

Will it add experience to my draft?

No. It lists where experience is missing and gives you questions for the author. Real experience comes from the person who has it.

Why does a team byline fail?

Readers and raters look for a person who stands behind the content. "The marketing team" can't have a bio, a track record or a reputation of its own.