AI SEO Governance Policy: A Template for Small Teams
A one-page AI SEO governance policy in ten clauses: what AI may and may not do, disclosure, sources, review, data, records and corrections, plus a risk register with owners. The Google, FTC and EU rules behind each clause are linked.
An AI SEO governance policy is a short written agreement on how your team uses AI to research, write and publish pages meant to rank in search: what is allowed, what is banned, who checks what and what gets recorded. Without one, every writer makes those calls alone, and the first time anyone reads the rules is after a page with an invented statistic or a fake testimonial has gone live.
Below: the policy as ten clauses to paste into a shared document, a risk register with scores and owners, the outside rules behind each clause and a one-week rollout. Example team: a five-person SaaS marketing team with two writers, an editor, a founder and a freelancer.
Summary
A small team’s AI SEO governance policy needs ten short clauses: scope, allowed uses, banned uses, disclosure, sources, review, data, records, corrections and ownership. The bans that matter most: no invented facts, no AI-written reviews or testimonials, no fake author profiles and no bulk template pages. Behind them sit Google’s spam policies, the FTC’s rule on fake reviews and Article 50 of the EU AI Act, which applies from 2 August 2026. Pair the policy with the risk register below and review both every six months.
Why does a small team need an AI SEO policy at all?
Because rules written far from your blog now reach straight into it:
- Google. Its spam policies demote or remove scaled, low-value pages whoever produced them, and its quality guidance treats fabricated author profiles as deception. The AI SEO writing guide quotes both.
- The FTC. Its final rule on fake reviews and testimonials, announced on 14 August 2024, covers reviews that misrepresent they are by someone who does not exist, “such as AI-generated fake reviews”, and lets the agency seek civil penalties against knowing violators.
- The EU AI Act. The European Commission lists its transparency obligations for AI-generated content as applicable from 2 August 2026.
- Copyright. The US Copyright Office concluded in January 2025 that AI output is protected only where a human author shaped it.
Add the everyday risks, such as a confident invented statistic or a customer’s email pasted into a free chatbot, and a team of five faces AI decisions every day. A one-page policy makes them once, and it stays one page because a long one never gets read.
What may AI do, and what is off limits?
Copy clauses 1 to 10 into a shared document and replace everything in square brackets.
Clause 1. Scope
This policy applies to every page [Company] publishes on [domain] and every guest article written in our name, whenever an AI tool (ChatGPT, Claude, Gemini, Copilot or any other) helped with research, outlines, text, images or metadata. It applies to staff and freelancers alike.
Clause 2. What AI may do
- Summarize sources that a team member chose and opened.
- Propose outlines, headings, title options and meta descriptions.
- Draft sections from an approved brief and source pack.
- Suggest edits, internal links and image alt text for a person to approve.
- Translate our own articles, for review by a fluent speaker before publishing.
Clause 3. What AI may not do
- Supply any statistic, price, quote, study, date or legal statement that a person has not confirmed at its source.
- Write reviews, testimonials, case studies or quotes presented as coming from real customers or experts.
- Create author names, photos, job titles or credentials.
- Produce batches of pages from one template with only keywords, places or dates swapped.
- Rewrite another site’s article as ours.
- Publish, schedule or update a live page without a named person’s sign-off.
Clause 3 traces to outside rules: bulk template pages and rewritten articles to Google’s scaled content abuse and scraping policies, invented authors to its guidance on fabricated creator profiles, fake testimonials to the FTC rule. Unconfirmed facts are banned for a plainer reason: they are wrong in public, under your brand.
How does the policy handle disclosure, sources and review?
Clause 4. Disclosure
Every article carries the byline of the person who signed it off, linked to an author page. Where a reader would reasonably ask how a page was made (data roundups, tool comparisons, translated pages, pages drafted mostly by AI), we add one line under the title: “Drafted with AI assistance from the sources linked on this page. Edited and fact-checked by [Name] on [date].” Text published to inform the public on matters of public interest for EU readers either carries an AI label or goes through clause 6 with a named editor responsible for it.
Clause 5. Sources and facts
Every statistic, price, quote and policy or legal statement links to the page where a team member read it, opened on or after the day the draft was written. Prices and plan limits come only from the vendor’s own pricing page and carry the date checked. Anything that cannot be confirmed is removed. AI output, AI search answers and AI summaries are never a source.
Clause 6. Review and sign-off
Each article gets two passes before publishing: an editor checks every claim against its source and against the brief, and the page owner signs off. Pages on health, money, legal or safety topics also need a reviewer with relevant qualifications. Titles, meta descriptions, structured data and alt text are reviewed with the same care as the body.
Where clause 4 comes from. Google’s generative AI content guidance says “Sharing information about how a piece of content was created can help give your readers more context.” Its helpful content page turns that into three questions for anyone using automation to generate content in bulk:
- “Is the use of automation, including AI-generation, self-evident to visitors through disclosures or in other ways?”
- “Are you providing background about how automation or AI-generation was used to create content?”
- “Are you explaining why automation or AI was seen as useful to produce content?”
The EU line comes from Article 50(4) of the AI Act: deployers of an AI system that generates text “published with the purpose of informing the public on matters of public interest shall disclose that the text has been artificially generated or manipulated.” The duty does not apply “where the AI-generated content has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication of the content.” A named editor under clause 6 is how a small team meets that exception. Reviewer roles and the fact-check log for clauses 5 and 6 are set out in human review for AI SEO articles.
How does the policy handle data, records and corrections?
Clause 7. Data and privacy
Nobody pastes customer personal data, contracts, unreleased financial figures, passwords or API keys into an AI tool. Other internal material goes only into tools on the approved list, with model training switched off wherever the plan allows it.
| Approved tool | Plan | Training setting | Account owner |
|---|---|---|---|
| [Tool] | [Plan] | [Off, as confirmed in the account settings on date] | [Name] |
Why the setting column matters: ChatGPT’s plans page (checked 5 Nov 2026) lists “Content is used to train our models” for its Free, Go, Plus and Pro plans, with an opt-out available. On consumer plans, training is the default until someone changes it.
Clause 8. Records
For each published article we keep an AI-use record:
| Field | Example entry |
|---|---|
| Article URL | /how-to-calculate-billable-hours/ |
| Tools and plan | Claude, Team plan |
| What AI did | Two outlines, first drafts of sections 2 to 5, five title options |
| Source pack | Link to the source document |
| Editor and date | [Name], 4 Nov 2026 |
| Sign-off and date | [Name], 5 Nov 2026 |
| Disclosure line shown | Yes, or no with the reason |
| Corrections after publishing | Link to the correction log |
The record shows a reviewer what to check and documents the human contribution. The US Copyright Office’s Part 2 report says AI-generated outputs “can be protected by copyright only where a human author has determined sufficient expressive elements”, and that protection does not extend to “the mere provision of prompts.”
Clause 9. Corrections
Reported errors are fixed within [2 working days]. When a correction changes the meaning of a sentence, the page gets a dated correction note at the end. Each correction is logged with its cause, and the same cause twice changes the brief template or the prompt.
Clause 10. Ownership and review
[Name] owns this policy. It is reviewed every six months, and sooner when Google changes its spam policies or AI content guidance, or when a law in one of our markets changes. Version [1.0], adopted [date], signed by everyone who writes, edits or publishes for [Company].
What goes in the risk register?
Ten risks a small team meets in its first year of AI-assisted publishing. Likelihood and impact run from 1 (low) to 3 (high), multiplied into a score; the scores are the example team’s, so rescore them for yours.
| Risk | Example | Likelihood × impact | Fix | Owner |
|---|---|---|---|---|
| Invented statistic or quote | A draft cites “73% of buyers” with no source behind it | 3 × 3 = 9 | Clause 5; [SOURCE NEEDED] markers in every draft | Editor |
| Outdated price or plan | A comparison lists a plan the vendor has renamed | 3 × 2 = 6 | Pricing page checked on publish day, rechecked every six months | Writer |
| Bulk template pages | 150 “[service] in [town]” pages from one prompt | 2 × 3 = 6 | Clause 3; owner approves any batch of more than 5 pages | Owner |
| Fake author profile | A freelancer invents “Dr. Jane Miles” with an AI headshot | 1 × 3 = 3 | Clause 4; author pages for real people only | Owner |
| AI-written testimonial | A landing page “customer quote” generated in a chatbot | 2 × 3 = 6 | Clause 3; written permission on file for every testimonial | Marketing lead |
| Customer data in an AI tool | Support tickets with names pasted into a free chatbot | 2 × 3 = 6 | Clause 7; approved tools list with training off | Owner |
| Copied competitor content | An AI paraphrase of the top-ranking article | 2 × 2 = 4 | Clause 3; outlines built from your own source pack | Editor |
| Wrong metadata | A meta description promises a free plan that does not exist | 2 × 2 = 4 | Clause 6 metadata review | Editor |
| Missing disclosure | An EU public-interest article goes out with no editor named | 1 × 2 = 2 | Clause 4 | Owner |
| Broken AI-suggested links | Links to merged or deleted pages | 2 × 1 = 2 | Clause 2 approval; monthly crawl for broken links | Writer |
How do you score the risks and decide what to fix first?
Multiply likelihood by impact for a score from 1 to 9, then sort:
- 6 to 9: fix this month.
- 3 or 4: fix this quarter.
- 1 or 2: accept, and recheck at the six-monthly review.
For the example team the register gives five risks to fix this month (invented statistic, outdated price, bulk pages, testimonials, customer data), three this quarter (copied content, metadata, fake author) and two to watch (missing disclosure, broken links). Start with the 9: it is the failure AI drafts produce most readily.
One score needs a human override. A fake author scores only 3, yet clause 3 bans it outright, because one discovered case undermines every other byline on the site; the fix is a two-minute check that each byline belongs to a real person.
Which outside rules does the policy answer to?
| Rule | Who it reaches | What it says | Clause |
|---|---|---|---|
| Google’s generative AI content guidance (updated 1 Oct 2026) | Any site that wants Google traffic | Fact-check AI output and its metadata by hand before publishing; give readers context on how content was made | 4, 5, 6 |
| Google spam policies | Same | Scaled content abuse, scraping, and site reputation abuse for third-party pages hosted mainly to borrow a site’s ranking signals | 3 |
| FTC rule on reviews and testimonials (announced 14 Aug 2024) | Businesses marketing to US consumers | Bans fake reviews and testimonials, including AI-generated fake reviews; civil penalties for knowing violators | 3 |
| EU AI Act, Article 50(4) (applies from 2 Aug 2026) | Deployers publishing AI-generated text to inform the public on matters of public interest | Disclose it, unless it was reviewed by a person and someone holds editorial responsibility | 4, 6 |
| Code of Practice on Transparency of AI-generated Content (final version published 10 June 2026) | Voluntary; about 190 organisations had signed by the end of July 2026 | Confirmed by the Commission and the AI Board as an adequate voluntary tool for showing compliance with Article 50 | 4 |
| US Copyright Office, Part 2 report (29 Jan 2025) | Anyone claiming copyright in AI-assisted text | Protection only where a human determined sufficient expressive elements; prompts alone are not enough | 8 |
Have a lawyer read clauses 3, 4 and 7 against the markets you sell into before you rely on them, and add any rule specific to your sector (health claims, financial promotions) as a line in clause 3.
How do you roll the policy out in one week?
- Day 1 (1 hour). The owner fills in the brackets and sets the six-month review date.
- Day 2 (1 hour). List every AI tool the team uses, free accounts included, switch training off where offered and fill in the clause 7 table.
- Day 3 (45 minutes). Team session: test clause 3 on three real drafts. Every “is this allowed?” question becomes a line in the policy.
- Day 4 (1 hour). Add the clause 8 record fields to the brief template or to custom fields in your CMS, so the record fills in as the work happens.
- Day 5 (2 hours). Audit the last 10 published articles against clauses 3 and 5, log what you find in the risk register and fix everything that scores 6 or more.
From then on, every new writer or freelancer signs the policy before a first brief. The per-article steps are in the AI SEO content workflow.
Which guides put the policy to work?
Each of these covers one clause in day-to-day detail.
what Google’s AI guidance and spam policies say, quoted from the source.
reviewer roles and a fact-check log for clauses 5 and 6.
the ten steps for one article, with the policy’s checks built in.
a scoring rubric to use alongside the sign-off in clause 6.
FAQ
Does a one-person blog need an AI SEO policy?
Yes, a short one. Clauses 3, 5 and 7 take ten minutes to write and decide what you do at 11pm before a deadline, and the clause 8 record documents your own contribution for copyright.
Does the EU AI Act make us label every AI-assisted blog post?
Article 50(4) covers AI-generated text published to inform the public on matters of public interest, and it does not apply where the text went through human review or editorial control with a person holding editorial responsibility. The article itself does not list which topics count, so clause 4 takes the safe path: a named editor on every article and a label on anything nobody reviewed.
Can we use AI to write testimonials, or to polish real ones?
Writing new ones is out: the FTC’s 2024 rule names AI-generated fake reviews among the practices it bans. For real testimonials, this policy allows edits for length only, and the edited wording goes back to the customer for approval before it appears on the site.
Who owns the copyright in an article drafted with AI?
Under the US Copyright Office’s January 2025 report, protection covers what a human author contributed: expressive elements they determined, or their creative selection, arrangement and changes to the output. Prompts alone do not qualify. Your clause 8 records show that contribution.
Can we paste customer emails into ChatGPT to find content ideas?
Only with names and identifying details removed, and only in a tool on your approved list. ChatGPT’s plans page shows content used for model training on its consumer plans unless you opt out, which is why clause 7 asks for the setting to be checked and recorded.
When should we review the policy?
Every six months, plus whenever a rule behind it changes. Google updated its spam policies page on 28 August 2026 and its generative AI content guidance on 1 October 2026, so changes arrive more than once a year.