AI content production

Where AI helps with outlines and summaries, why it cannot be trusted on facts, figures and quotations, and a workflow built on a writer, source checks and a separate editor.

· 11 min readContent

Short answer

Using AI for content is safe when the tool is an aid and a writer is accountable. Google does not treat appropriate use as against its guidelines; the problem is generating many pages without adding value to manipulate rankings. Check every fact at its source and have someone other than the writer read the text.

A draft lands on the editor’s desk. The language is fluent and the headings are tidy, but the second paragraph opens “research shows that most customers…” and the text never says which research. In the third section, an expert is quoted in quotation marks; when someone asks the expert, it turns out they never said it. This is where AI content production most often goes wrong: the text arrives quickly, but nobody has checked what it says against its source. In this article we cover where the tool genuinely helps, where it should not be trusted, and a seven-step workflow in which a writer and an editor step in.

What Google says about AI-generated content

Google Search Central’s guidance on generative AI opens with two points. Generative AI can be particularly useful when researching a topic and to add structure to original content; but using it to generate many pages without adding value for users may violate Google’s spam policy on scaled content abuse. The spam policy gives the same example: using generative AI tools to generate many pages without adding value for users.

Three things follow from the guidance. First, Google looks at the quality of the content and does not stop at the method of production. The Search Central blog says that appropriate use of AI or automation is not against its guidelines, and that what breaks the rules is generating content primarily to manipulate search rankings. Second, the guidance notes that generative models don’t retrieve facts but predict a likely sequence of words, so outputs can contain inaccuracies (hallucinations), and it calls it critical to manually fact-check and review all AI-generated content before publishing. That review covers the title element, the meta description, structured data and image alt text too. Third, saying how a piece of content was made gives readers context; Google frames this as “who, how and why” and suggests adding disclosure where readers would reasonably expect it. It also leaves no room for deception about authorship: AI-generated headshots, made-up names or false credentials used to make content look as if human experts wrote it count as deception.

Detmore’s position on content work points the same way: AI is an aid, every text has a writer who is accountable for it, and no text is published without being read by someone other than its writer.

Where AI helps and where it can’t be trusted

The line depends on what the tool is doing. We cover how to steer the tool on your own tasks, the parts of a prompt and how to check the answer in how to write prompts for work. On structure and format it saves time; on anything that is simply right or wrong, such as a fact, a figure, a quotation or a legal statement, its output is a claim, not knowledge. The table below sorts the jobs.

Job Can it be left to AI? What you do before publishing
Topic outline, headline and section-order suggestions Yes, as suggestions The writer sets the order from the reader’s question
A rough draft from your own notes Yes, only from notes you supply The writer rewrites it in their own sentences
A summary of a long document or an interview Yes Compare the summary with the source; look for gaps and additions
Reformatting (bullets to paragraphs, text to a table) Yes Read it to confirm the meaning has not changed
Numbers, dates, percentages No Open the source and take the figure from there
Quotations and words attributed to people No Confirm with the person or the original published text
Legal and regulatory statements No Read the official text; ask your own lawyer if needed
Product and service features No Ask the product owner or check the official documentation
Lists of sources and links No Open every link; check the page exists and contains the claim

What the “yes” rows share is that the yardstick for correctness is already in your hands: notes, a document, a recording. The tool arranges that material. The “no” rows are where a tool steps outside its material and fills the gap itself.

The seven-step workflow

  1. Gather your own material first. Write the reader’s question in one sentence, then put under it your notes, the recording of the expert interview, the product information and the sources you will open. A text that starts without material is built from the tool’s general knowledge and has nothing new to say for your site. We described how to run the expert interview in the B2B content plan article.
  2. Use the tool for aid work. A section-order suggestion, a rough draft from notes, a summary of a long document, turning bullets into paragraphs: the “yes” rows of the table. Give it only the material that job needs (see the section below), and treat what comes back as raw material rather than a finished text.
  3. The writer builds the text in their own sentences. The writer writes the sentences that reach the reader. Leaving a paragraph from the tool with a few words swapped does not make it yours; the writer rebuilds each sentence around whether it answers the reader’s question. Having a tool reword another site’s text can also fall under the copying that Google’s spam policy describes.
  4. Check every fact at its source. List every number, date, name, quotation, product feature and legal reference in the text, and next to each put the source you opened. If the tool named a source, open that too, because it falls to you to make sure the page exists and contains the claim. A sentence you cannot find in a source comes out. The same check applies to the title element, the meta description, structured data and image alt text.
  5. Someone other than the writer reads it. Writers rarely see the gaps in their own sentences, so the editor’s job is a second reading: structure, style, final read. The editor should see the source list too, because answering “is this true” requires knowing where the claim came from. At Detmore, no text is published without being read by an editor other than its writer.
  6. Decide whether to disclose. Google suggests telling readers when automation, including AI, was used to substantially generate content, and adding that where readers would reasonably expect it. If you used the tool only for summaries and formatting and the writer built the text, the writer’s name is the main thing to state. For images or product data, platforms can have their own rules; Google says Merchant Center requires a specific metadata field for AI-generated images and separate labelling for AI-generated product data.
  7. Keep a record. For each text: who wrote it, who edited it, which sources were checked, where the tool was used. Three lines are enough. When the text is updated a year later, or a reader questions a claim, the answer is in that note.
The AI-assisted content workflow: from your own notes to the writer, the source check, a separate editor and publication 1 Your own notes and sources 2 AI as an aid to the writer 3 The writer rewrites it 4 Every fact checked at source 5 A different editor reads it 6 Publish, disclose and keep a record The AI-assisted content workflow: from your own notes to the writer, the source check, a separate editor and publication 1 Your own notes and sources 2 AI as an aid to the writer 3 The writer rewrites it 4 Every fact checked at source 5 A different editor reads it 6 Publish, disclose and keep a record
Figure: AI is an aid at step two only; the writer writes the text, every fact is checked at its source, and someone other than the writer reads it before publication.

What not to paste into an AI tool

Whatever you give a tool goes into another company’s system. Türkiye’s data protection authority (KVKK) published a generative AI guide in November 2025. It advises users not to share personal data that identifies a person, such as a name and surname, address, telephone number or identity details, and to take care not to share other people’s personal data. It advises against sharing health, financial and legal-process information, and, where sharing is necessary, using anonymised, general wording. It also reminds readers that transferring personal data through a service provider based abroad must comply with Article 9 of Law No. 6698 and the related regulation.

In practice, this means keeping the following out of the tool:

  • Customer names, email addresses, telephone numbers and order records.
  • Unpublished contracts and proposals.
  • Any record about an employee, a candidate or a patient.
  • Account passwords, access keys and internal system details.
  • Confidential documents that belong to a customer, without that customer’s written permission.

Where needed, swap names and details for general wording such as “a customer” or “a product” before you paste. How a tool stores and processes data is written in its own privacy notice and privacy policy; read it before you build a workflow that touches personal data, and ask your own lawyer. This article is not legal advice.

Common mistakes

  • Publishing the tool’s output. A fluent text is not a correct text; Google’s own guidance says generative models don’t retrieve facts and can produce inaccuracies.
  • Asking the tool for a source and not opening it. It may give a link or a publication name, and either may not be real. A source you have not opened has not been checked.
  • Making the writer and the editor the same person. A person reading their own text remembers what they meant to say and sees that rather than what is on the page. The second pair of eyes has to belong to someone else.
  • Mass-producing pages. Google’s spam policy names generating many pages without adding value as an example of scaled content abuse. A few pages that answer a question readers really ask usually do more work; for choosing topics, see how to write SEO-friendly content.
  • Using a fake author or expert profile. A made-up name, an AI-generated photograph or a title that does not exist is described as deception in Google’s guidance and makes the page untrustworthy. The author should be a real person whose name leads to a real page; we covered how this relates to E-E-A-T and to AI answers in what is GEO. Our guide to what E-E-A-T is and how to show it on a site walks through trust signals such as author pages, sources and dates.
  • Pasting customer data into the tool. Pasting a real customer email to get a better draft opens that data to another company’s system.
  • Letting the tool write quotations and case results. A customer’s words and a project’s results come only from a real source the customer has approved. We explained how in how to write a case study.

Common questions

Does Google penalise content written with AI?

Google Search Central says appropriate use of AI or automation is not against its guidelines. What is against them is producing content primarily to manipulate search rankings, and its example is using generative AI tools to generate many pages without adding value. So the test is whether the resulting page adds value for the reader, not the tool itself.

Do I have to tell readers I used AI?

Google suggests adding disclosure where automation substantially generated the content and where readers might reasonably ask how it was made; this is a suggestion that gives readers context. If the writer built the text and the tool only helped with structure or summaries, the author’s name and an editorial process are usually enough context. Platforms (Merchant Center, for example) and contracts can have their own rules, so check those separately.

Can I give an AI tool customer information?

The KVKK guide advises against entering names, contact and identity details, and health, financial and legal-process information into these systems, and suggests anonymised, general wording where sharing is needed. If the provider is based abroad, transferring personal data is also subject to Article 9 of Law No. 6698. When in doubt, leave the data out and ask your lawyer.

One small job you can do today: open the last text an AI tool helped with, list every number, date, name and quotation in it, and put next to each the source you opened. The lines where you cannot find a source are the places to fix at the text’s next read.

Sources

This article is general information, not legal, financial or technical advice. Platforms, their rules and the law change over time, so some details may have changed since 11 November 2026 or contain small errors. Before you make a decision for your business, check the current official sources or write to us.

Tell us about
your project.

Fill in the form or write to us directly, and we will agree on a day to talk.

Your first meeting is with Deniz Sezer, founder of Detmore.

First meeting free · Reply within one working day · Written proposal in 2 to 3 days · NDA ready

What do you need? pick as many as you like

Name, email and phone are required.

Contact

We reply within one working day.