AI and content
AI makes it faster. You make it better.
AI researches, writes first drafts, creates images, translates and produces variants for each channel. That makes content cheaper. But what anyone can produce in minutes sets nobody apart. Here you can read what AI does in every phase, how much freedom you give it, how five approaches compare and how to get cited in AI answers.
Figures
Creating faster is not yet performing better.
92.4%
of content marketers use AI. Among B2B marketers, 95 percent of organisations use AI applications.
Source: Orbit Media 2026 (n = 1,042) and CMI 2026 (n = 1,015)
What it does not prove That AI leads to better results. According to Orbit Media, users and non-users are equally likely to see strong results.
87%
say AI improved productivity. Only 39 percent say AI improved the performance of their content.
Source: Content Marketing Institute, B2B Content and Marketing Trends 2026
What it does not prove That AI harms performance. It shows that creating faster is not the same as performing better.
1%
of Google visits with an AI summary lead to a click on a link in that summary. 8 percent click on a regular result, compared with 15 percent without a summary.
Source: Pew Research Center, July 2025, 68,879 searches
What it does not prove That Belgian users behave the same way. The panel consisted of 900 American adults.
AI per phase
What AI does, from question to measurement.
AI helps in every phase of your content, but not equally well everywhere. For each phase you see what AI does, where to start and what to watch out for.
- 01
Research
What AI does Summarising customer questions, reviews and sales conversations, grouping search queries, comparing competitors.
Start with Let AI summarise the transcripts of ten customer conversations into questions and objections.
Watch out Made-up sources and figures. Check every source yourself.
- 02
Briefing
What AI does Filling in a briefing with goal, audience, job, structure and sources, based on your template.
Start with A fixed briefing template that AI completes each time and a human approves.
Watch out A briefing without your own angle. Then you get what everyone writes.
- 03
First draft
What AI does Writing a first draft in your voice, based on the briefing, your own knowledge and examples.
Start with A flow with your tone of voice, your best pieces as examples and a list of banned words.
Watch out Smooth, generic text. Let AI start from an interview or your own notes, not from nothing.
- 04
Images and video
What AI does Diagrams, illustrations, thumbnails, subtitles and short clips from long videos.
Start with Clips and subtitles from your latest webinar.
Watch out Images that do not show what you really do. Real photos of people and projects win trust.
- 05
Translation
What AI does Translating and localising into French, English or German, with your own glossary.
Start with Your best-performing piece in a second national language, proofread by a native speaker.
Watch out Literal translations and Netherlands Dutch words in Flemish text. A native speaker always proofreads.
- 06
Final editing
What AI does Flagging facts, figures and names to check, improving readability, guarding consistency.
Start with A checklist that AI runs through before the human does the final read.
Watch out Letting AI decide what is correct. Final responsibility lies with a human.
- 07
Distribution
What AI does Variants per channel: posts, newsletter, ad copy, summaries for colleagues.
Start with Five derivatives per core piece, in the voice of each channel.
Watch out The same text everywhere. Every channel and every colleague deserves their own angle.
- 08
Measurement
What AI does Summarising figures from analytics and CRM, looking for patterns, suggesting updates.
Start with A monthly summary of which pieces appeared in leads and deals.
Watch out Reading correlation as cause. Let AI suggest hypotheses, and test them.
How much freedom
Not every task deserves the same freedom.
The bigger the consequences of a mistake for your brand, your customer or the law, the closer the human stays to the decision.
- Human decidesAI helps, you decidePositioning, promises, prices, figures and claims, legal and sensitive topics, anything that makes a promise on behalf of your company.
- Human editsAI creates the first draft, a human finishes itArticles, case studies, newsletters and scripts. A human checks facts, adds experience and does the final edit.
- AI does, human checksAI does the work, a human spot-checksVariants per channel, summaries, metadata, alt texts, subtitles and internal notes.
Approaches compared
Five ways to create content with AI.
From prompting yourself in a chat window to a flow built around your business. Each approach has its place.
| Approach | What it is | Strong at | Watch out |
|---|---|---|---|
| Prompting yourself | ChatGPT, Claude or Gemini in a chat window, one prompt at a time. | Quick starts, ideas, first drafts and rewriting. | Explaining who you are every time. The result depends on who is prompting. |
| AI writing tools | Ready-made tools with templates for blogs, posts and ads. | Producing lots of standard copy quickly, with a team. | Generic output that resembles your competitor's. Your knowledge is not in it. |
| Tailored AI flows | A fixed series of steps around your knowledge, voice, templates and checks, with a human in the right places. | The same quality every time, faster and cheaper, in your voice. | Requires a set-up and an internal owner who uses and improves the flow. |
| Agents | AI that carries out several steps independently: researching, writing, publishing, reporting. | Repetitive work with clear rules and checks. | Who checks what gets published? Grant permissions step by step. |
| Agency with AI | A partner that uses AI in its own process and delivers you the result. | Rhythm and quality without your own time. | Ask where AI is used, what a human does, and who owns the flows and prompts. |
A briefing for AI
Good input, better content.
AI is as good as the briefing you give it. The same briefing also helps your colleagues, your freelancers or your agency. Put it in a template and use it every time.
- The goal and the reader: who are they, in what situation, and what should they do after the piece?
- The reader's job in one sentence, in their own words.
- Your own raw material: interview, notes, figures, examples from projects.
- Your voice: three examples of pieces you like, and words you never use.
- The structure: answer first, subheadings, length, and the next step at the bottom.
- The checks: which facts, figures and names a human must verify.
Getting cited
Ten things to get cited by ChatGPT, Gemini and Perplexity.
- 01
Answer real customer questions in plain language, with the answer in the first two sentences.
- 02
Use subheadings that read as a question or a statement.
- 03
Give figures with source and date. They are the easiest to pick up.
- 04
Publish something only you have: your own data, experience, examples from projects.
- 05
Put an author with name, job title and experience under every piece.
- 06
Make sure the content is on the page as plain text, not only in a PDF, an image or behind a script.
- 07
Keep your pages up to date and show when they were updated.
- 08
Get others to write about you: trade media, partners, reviews. AI cites what is frequently confirmed.
- 09
Regularly ask AI assistants what they say about your topic, and note whom they cite.
- 10
Measure mentions and recognition too, not just clicks.
Pitfalls
Five ways AI makes your content worse.
- 1
Made-up facts
AI always sounds certain, even when a figure, a source or a quote does not exist. Every factual claim gets checked.
- 2
Interchangeable text
Without your own knowledge, AI writes what everyone writes. Mediocrity at scale is still mediocrity.
- 3
Volume over value
Because you can, you make more. But more pieces nobody needs dilute your site and your brand.
- 4
No owner
If nobody manages the flow, errors creep in and the voice disappears. A flow needs maintenance.
- 5
No transparency
Since 2 August 2026, the transparency rules of Article 50 of the AI Act apply. Disclose AI where required and where your audience expects it.
FAQ
Questions about AI and content.
Does Google penalise content created with AI?
No. Google judges content on quality and usefulness for people, not on how it was made. Content made mainly to rank, with or without AI, does perform worse.
How much time does AI really save?
At Orbit Media, the time for an article fell from 4 hours 10 minutes in 2022 to 3 hours 20 minutes in 2026, while AI use rose to 92.4 percent. That is a correlation. With a tailored flow, an honest time saving is usually between 30 and 60 percent, because a human keeps checking and editing.
What is a tailored AI flow?
A fixed series of steps, built around your knowledge, your voice, your templates and your checks: from briefing through research and first draft to variants per channel, with a human in the places where it counts. The same quality every time, faster than before.
Do I have to disclose that content was created with AI?
Article 50 of the AI Act has applied since 2 August 2026. Transparency obligations apply to AI-generated text on matters of public interest, and to deepfakes. For ordinary marketing copy edited by a human, disclosure is usually not mandatory, but it is honest. Seek legal advice if in doubt.
Can AI learn my tone of voice?
Largely, if you give it good examples, write down your rules and have a human do the final read. The voice lies mainly in what you think and say, and that has to come from your own knowledge.
Do AI answers make content worthless?
No. AI can only cite what someone has created first. But the value shifts: from clicks to mentions and recognition, and from general explanations to your own data, experience and answers you cannot find anywhere else.
Want to know what an AI flow would save you?
Work out what a tailored flow saves per year, or first find out which framework suits you.
Finally: the harness
With a harness you get even more out of Claude Cowork.
Claude Cowork is strong on its own. But the result depends not only on the model, but also on the layer around it: the harness. It gives the model a workplace with your company's knowledge, roles, methods and tools, so every colleague works with the same memory and nobody has to explain again how you do things. For content, that means your voice, your sources and your flows are built in. Edge has such a harness: Zaela. This site was built with it too.
- The loopgather context, work, review, repeat
- Toolsaccess to your data, systems and apps
- Memorybrand, customers and previous work in one place
- Permissionswhat the agent may do on its own, and what not
- Oversighta human keeps an eye on things where it counts