- Is it worth publishing AI content without human review?
- What Ahrefs research really says: the numbers that matter
- What are the concrete risks for your company if you don't review AI content?
- How does human review that truly protects work in practice?
- Three concrete scenarios: how does the situation change for an Italian company?
- What tools help manage the review of AI content?
- Why are people talking about this now and not a year from now?
- How to build a review process: concrete steps to get started
- What are the most common mistakes in managing AI content?
- How to tell if the review process is working?
- So, is it worth structuring a review process for AI content?
- How we see it at SHM
- Frequently Asked Questions
- 93% of marketers already use AI to create content, but 89% maintain human review: supervision is still the norm, not the exception.
- Publishing AI texts without editing exposes you to three real risks: loss of Google rankings, fines up to 15 million euros for the European AI Act, and a drop in public trust.
- Building a structured review process — not just a simple reread — is the most important skill today for those using AI in marketing.
- 76% of Italian SMEs have not yet invested in AI: those who start now can do so with the right rules from the beginning.
7% of marketers using AI publish texts as they come out of the tool, without touching them. A year ago, it was 4%. If your company produces content with AI, this trend directly concerns you — even if you're not among that 7%.
Is it worth publishing AI content without human review?
No, almost never. The risks — reputational, ranking, and legal — far outweigh the time saved. Human review remains the main quality control tool, even among those who use AI every day.
Ahrefs' annual survey of 300 marketing professionals, published on September 30, 2026, paints a clear picture: the 93% of marketers use AI to create content , but the 89% check manually what AI produces before publishing it. Only 7% publish AI texts without significant changes.
That 7% has grown compared to 4% in 2025. It's not yet the majority, but the direction is visible. And the consequences of that choice — for the brand, for rankings, and for compliance with new European rules — are concrete.
| Your situation | Is it a good idea to publish without reviewing? | why |
|---|---|---|
| You publish content on technical or industry topics | No, for now | AI gets the details wrong: a technical error damages credibility more than a publication delay |
| You produce corporate or brand content | No, for now | The brand's tone and voice cannot be replicated without human intervention |
| You use AI for internal drafts or brainstorming | Yes | For internal use, formal review is not necessary |
| Publish on topics of public interest (health, finance, law) | No, for now | The European AI Act requires labeling or substantial editorial review |
| You have a dedicated editor reviewing every piece | Yes | With structured review, AI accelerates without lowering quality |
What Ahrefs research really says: the numbers that matter
The most striking data point isn't the 7% who don't review, but the fact that 89% keep human review even as trust in AI grows. Speed goes up, oversight doesn't drop.
Ahrefs interviewed 300 marketing and SEO (Search Engine Optimization) professionals in August 2026. The results show a market that is using AI more and more intensively, but which has not yet abandoned human control. You can read the complete analysis of the source at PPC Land .
One data point is worth noting: 29% of marketers use an AI to evaluate texts produced by another AI — a practice that has doubled from 14% in 2025. The problem is that this approach doesn't replace human review. As a content marketer from Ahrefs who tested the method observed: even when the AI flagged issues, the feeling was that you didn't know what else it had gotten wrong or missed.
Speed is the clearest advantage: 85% of marketers produce content faster, and among executives the share rises to 62% who report working "much faster." But speed without strategy, as an Ahrefs SEO educator put it, "tastes like a factory."
What are the concrete risks for your company if you don't review AI content?
Three distinct and measurable risks: a drop in Google rankings, legal penalties of up to 15 million euros, and a loss of public trust that a recent study quantified as doubling in just one year.
The risks are not theoretical. They concern three separate areas, each with concrete consequences for the business.
| Risk area | What happens | Extent of damage |
|---|---|---|
| Google ranking | Google's quality rater guidelines (updated January 2025) classify pages with automatically generated main content without original added value as 'Lowest' — the worst score. | Loss of organic visibility, traffic, and leads |
| European AI Act | Starting August 2, 2026, anyone publishing AI texts on matters of public interest without substantial editorial review must label them as AI-generated. | Fines up to 15 million euros or 3% of global turnover |
| Brand trust | According to a Fractl study (2026, 1,008 US consumers), the share of people who say that heavy AI use would reduce their trust in a favorite brand has doubled: from 20% in 2025 to 40% in 2026 | Reputation erosion, increased churn |
On the legal front, the article 50 of the AI Act (the EU AI Act) provides an important exemption: if the AI text has undergone human review substantial by those who assume editorial responsibility, the labeling obligation does not apply. The review must include fact-checking, not just spell-checking. You can read the details of the European Commission guidelines at Altalex and operational deadlines on CityNext .
To learn more about how Google evaluates AI content, an analysis of the Quality Rater Guidelines is available at Fullstackoptimization .
How does human review that truly protects work in practice?
Just proofreading the text isn't enough: the review that matters is the one that adds value — fact-checking, brand voice, original examples. That's what satisfies both Google and the AI Act.
There is a clear difference between a formal review and a substantial review. The first is a spelling and readability check. The second transforms AI text into content that bears the company's editorial signature.
- 1 · Writer or internal expert Fact-checking Check every data point, statistic, and technical claim. AI often invents plausible but false numbers (so-called 'hallucinations').
- 2 · Copywriter or marketing manager Brand tone adaptation Rewrites parts that sound generic. Adds concrete examples, anecdotes, or references to the Italian market.
- 3 · Industry expert (if available) Technical review For specialized content (legal, medical, financial, technical), an expert verifies substantial correctness.
- 4 · Editor or content manager Editorial approval Whoever signs off on the content takes on editorial responsibility. This step is what triggers the exemption from the AI Act labeling rule.
- 5 · SEO or marketing team Optimization and publishing Adds meta tags, internal links, optimized images. Only at this point does the content go online.
Three concrete scenarios: how does the situation change for an Italian company?
Time savings are real, but depend on how the process is structured. Without review, savings turn into hidden costs.
Illustrative examples: numbers are meant to show the reasoning, they are not real customer data.
| Company type | Current use of AI | Risk without review | Recommended approach |
|---|---|---|---|
| Fashion e-commerce (20 employees) | AI-generated product pages and category descriptions | Duplicate or generic content penalized by Google; loss of organic traffic | Quick review (10 min/piece) to add original details and brand tone |
| Professional firm (accountant, lawyer) | Blog posts on tax or legal regulations | Factual errors on regulated matters; high reputational and legal risk | Mandatory review by an expert professional before each publication |
| B2B manufacturing company | Data sheets, case studies, trade show content | Incorrect technical data that compromises credibility with buyers | Technical review by the technical office + marketing approval |
What tools help manage the review of AI content?
There are tools for every stage of the process: from generation to verification, up to editorial management. The choice depends on content volume and team structure.
| Phase | Tool / approach | What it does | Limit to keep in mind |
|---|---|---|---|
| Generation | ChatGPT, Claude, Gemini, Ahrefs Letaido | Produces drafts, outlines, text variations | Does not fact-check: output must always be checked |
| Fact check | Manual research + primary sources | Check every single data point cited by the AI | There is no 100% reliable automatic tool |
| Plagiarism/originality check | Copyscape, Originality.ai | Check that the text is not copied from other sources | It doesn't spot factual errors or hallucinations |
| AI review on AI | Review features built into AI tools | Flags stylistic or consistency issues | It doesn't replace human review; it can give false confidence |
| Editorial management | Notion, Trello, Asana with approval workflows | Track who reviewed and approved each piece of content | Requires team discipline to be effective |
Why are people talking about this now and not a year from now?
Three factors converge in 2026: AI adoption in marketing teams has exploded, European regulations have come into force, and consumer trust in AI content is declining. Those who wait lose ground on all three fronts.
The Italian artificial intelligence market has reached 1.8 billion euros in 2025 , with a growth of +50% compared to the previous year, according to the Artificial Intelligence Observatory of the Polytechnic University of Milan . Generative AI — the kind that produces text, images, and videos — accounts for 46% of the total.
Yet, according to the Digital Innovation Observatory in SMEs of Politecnico di Milano , the 76% of Italian SMEs have not invested or plan to invest in AI , and only 7% have started structured training programs. Those who start now can build the right process from the very beginning, without having to fix bad habits.
At the European level, according to the report by IAB Europe on the impact of AI in digital advertising (2026) , 86% of organizations already use AI for marketing and 78% have identified a formal AI governance manager — someone who oversees how AI is used and controlled. In Italy, this role is still rare in SMEs.
How to build a review process: concrete steps to get started
We start with an audit of what you're already publishing, figure out who reviews what, and build a workflow that scales with your content volume.
The issue of AI content governance intersects with the broader editorial strategy. If you are considering how to position your content in AI search engines as well, you can dive deeper into the topic at Marketing and ADV performance , our cluster dedicated to paid media, content, and performance.
- 1 · Marketing manager or content lead Audit existing content List all the channels where you publish content (blog, social media, product pages, newsletters). Identify which ones already use AI and with what process.
- 2 · Marketing team Define risk levels Classify content by risk: high (technical, legal, health topics), medium (brand content), low (internal use, brainstorming).
- 3 · Content manager + legal (if necessary) Write the internal guidelines Establish who can use AI, for what content, and who must approve before publication. One page is enough to get started.
- 4 · Marketing team Choose your tools and workflow Use a project management tool (even just a shared spreadsheet) to track the review status of each piece of content.
- 5 · Entire team Train the team on the process Even an hour of internal training cuts down on mistakes. The 7% who publish without reviewing often do it due to a lack of processes, not by choice.
- 6 · Marketing manager Monitor and update Review the guidelines every quarter. AI evolves rapidly: what works today might not be enough in six months.
What are the most common mistakes in managing AI content?
The most frequent mistake is not using AI, but using it without a defined process. The second is believing that AI review on AI is sufficient.
The issue of trust in AI content also emerged in the analysis of Creator content and AI: more than half of consumers prefer them — a sign that the public already distinguishes between authentic and generic content. And if you are evaluating how AI is changing the relationship with your audience, also on the advertising front, it's worth reading the analysis on ChatGPT Ads click-to-chat and conversion without site traffic .
| Common mistake | Why it is a problem | How to avoid it |
|---|---|---|
| Publishing without a defined process | Each team member independently decides how much to review, with inconsistent results | Write internal guidelines, even brief ones |
| Using AI to review AI | AI doesn't detect its own hallucinations; it gives false confidence | The final review must always be human |
| Treating all content the same way | A social post and a technical white paper have very different risks | Classify content by risk level and adapt the process |
| Not updating the guidelines | AI capabilities change rapidly; yesterday's rules may not be enough | Quarterly review of internal guidelines |
| Ignoring the AI Act labeling requirement | In effect since August 2, 2026: not knowing the rule does not exempt from responsibility | Check if your content falls within the scope of Article 50 |
| Do not train the team | The 7% who publish without review often do so out of habit, not choice | Even a one-hour training session changes behaviors |
How to tell if the review process is working?
Three indicators are enough to monitor process health: perceived content quality, organic ranking, and the error rate detected during review.
| Indicator | How to measure it | Red flag |
|---|---|---|
| Rate of errors caught during review | Count the factual errors found before publication each month | If it increases, AI is producing less reliable content or the briefing is insufficient |
| Organic ranking of AI-assisted content | Monitor Google rankings for content produced with AI | A systematic drop signals quality issues perceived by Google |
| Average review time per piece of content | Track how much time each reviewer spends | If it exceeds writing time, the briefing to AI needs improvement |
| Audience feedback | Comments, shares, requests for clarification | An increase in questions about technical details signals superficial content |
| AI Act compliance | Verify that all content on topics of public interest has documented review | Any content published without a trace of review is a legal risk |
So, is it worth structuring a review process for AI content?
Yes, for almost all companies already using AI for content. The question isn't whether to do it, but how to structure it based on volume and risk.
It is worth it if you already use AI to produce content published on the site, blog, or social channels, and you don't yet have a formal review process. The legal and reputational risk justifies the investment even for small teams.
A middle ground produce AI content only for internal use or low-risk channels (informal social, short newsletters). A light process—even just a checklist—is enough for now.
Not yet if you don't use AI for content yet, or you only use it for brainstorming and outlines that are never published directly. You can wait, but it's best to prepare before scaling.
How we see it at SHM
For us, AI is a tool that gives back time — not one that replaces judgment. We believe the value of content lies not in the speed of its production, but in its ability to answer a real reader's question with precision and authority. This doesn't change with AI: only who does the routine work and who does the work that matters changes.
When we work with companies that want to integrate AI into their editorial processes, the first step isn't choosing the tool: it's understanding which content truly requires a human touch and which can be delegated. Consider a B2B manufacturing company that produces technical data sheets in five languages: AI can generate the structure and the first draft, but the technical review by an engineer and editorial localization (not simple translation) are the value that no tool can automatically replicate.
If you want to understand how to structure this process in your company, you can start with a consultation on AI Strategic AI integration , or explore how to optimize your content for new AI engines with our service of SEO for AI . For those who want to train their internal team, the path of AI training for businesses is the most practical starting point.
The topic of AI content governance also intertwines with visibility: if your content is not cited by AI search engines, it's worth reading the analysis on ChatGPT Ads in Asia and how to evaluate it for your strategy , and on Amazon DSP on ChatGPT Ads and opportunities for Italian B2B .
Frequently Asked Questions
It depends on the type of content and how much you've reviewed it. From August 2, 2026, the European AI Act (Art. 50) requires AI texts on topics of public interest to be labeled if they haven't undergone substantial editorial review. If you have an editor who fact-checks and takes responsibility for the content, the labeling requirement doesn't apply. For standard commercial content (product descriptions, social posts), the rules are less strict, but transparency is still recommended.
Google doesn't penalize AI as such, but it does penalize low-quality content, regardless of how it was produced. Google's quality rater guidelines (updated January 2025) classify pages with automatically generated content without original added value as 'Lowest'. AI text that has been reviewed, enriched with real-world examples, and fact-checked is treated like any other quality content.
It depends on the content's complexity. For an 800-word blog post on a non-technical topic, substantial review takes about 20-30 minutes on average: fact-checking, tone adaptation, adding original examples. For technical or legal content, the time can double. In any case, it's much less than the time needed to write the text from scratch.
No, and the data confirms it. 29% of marketers already use this approach, but those who have tested it report that AI does not detect its own hallucinations (i.e., invented but plausible information). The final review must always be human, at least for fact-checking. AI can help check style and consistency, but it does not replace editorial judgment.
Internal-use content — brainstorming, outlines, first drafts, document summaries — can be managed with AI without formal review. For published content, zero risk doesn't exist, but it's very low for short, low-impact formats (e.g., social media captions on non-controversial topics). The risk increases with technical complexity, legal relevance, and content visibility.
The AI Act art. 50 applies to AI-generated texts on 'matters of public interest' — a category that includes health, finance, law, politics, but can extend to other sectors based on interpretation. If you publish content on regulations, financial products, health, or safety, it's wise to consult a lawyer. For all other commercial content, substantial editorial review is still the safest bet.
Human review isn't an extra cost compared to AI: it's the step that turns a generic text into content that carries your company's name.
Related articles
Discover more articles exploring similar topics, selected to offer you a more complete and stimulating perspective. Each piece of content is carefully chosen to enrich your experience.
