One of the toughest editors I’ve had in years was an AI chatbot. I did all the outlining, drafting, and rewriting myself. AI offered research and editing suggestions, and I reviewed them and made every change myself.
For smart authors and organizations using thought leadership as a strategic way to grow their business, AI can be both a help and a hindrance. Publishing AI-generated writing exposes a brand’s voice, reputation, and legal protections to risk in an age in which AI writing can be spotted with watermarks and detectors.
Authors who use AI entirely to write make their books and articles ineligible for copyright protection, and many reputable publishers refuse to publish AI-generated text. Making matters worse, a growing amount of research into “cognitive atrophy” from MIT Media Lab and others shows that using AI to write from scratch without extensive human involvement threatens to degrade our collective critical-thinking abilities.
But there are ways to use AI to write, without letting it write.
A Case Study in Human-Led Content
An AI company recently hired me to write thought-leadership explainers for its blog to help readers understand AI agents and security. You might expect tech companies to be the first to replace writers and editors with AI. A few of them hire me instead. This client’s editors told me they wanted accessible, human-led writing with an original point of view in a field crowded with cookie-cutter technical explainers.
I researched, fact-checked, and wrote each article for a nontechnical audience, and I structured each one to be both rich in information and easily digestible, as well as to perform well on search engines and answer engines, which are closely related. At one point, the first article I wrote ranked seventh in organic Google results for its target keyword, alongside well-established tech participants in the conversation, an encouraging data point. Answer-engine results are not yet in.
I adapted the following process for thought-leadership content from a B2B content workflow by writer Pawel Tatarek, with input from the client’s content manager. The steps should resemble many traditional thought-leadership publishing workflows. What differs is where AI enters the process and where it stays out.
The Steps to Producing Human-Led Content
| Step | Who leads | What AI does |
| 1. Analyze what already ranks | Human directs AI | Runs a SERP analysis from reusable instructions |
| 2. Find an angle competitors are missing | Human | Makes suggestions in the SERP |
| 3. Give AI a specific research brief | Human | Organizes research |
| 4. Have a human read the sources | Human | Summarizes output, which a human reads, verifies, and supplements with additional research |
| 5. Have a human build the outline | Human | Nothing |
| 6. Let AI organize the research | Human directs AI | Sorts research into outline sections |
| 7. Have a human write the draft | Human | Nothing |
| 8. Encourage AI to critique the draft | AI provides input, a human decides | Flags logic gaps, unsupported claims, factual errors, and writing problems |
| 9. Rewrite with human editors | Human | Repeats the critique after each draft |
1. Analyze what already ranks
The client team and I built an AI “skill,” or a reusable set of instructions, to run a SERP (search engine results page) analysis of what search algorithms reward. For a target keyword, the analysis shows what currently ranks and suggests ways to structure and differentiate an FAQ-style article. It was a starting point for much more human refinement.
2. Find an angle competitors are missing
The analysis demonstrated the sameness of most ranking articles. Many tech companies publish explainers on the same topics to try to attract website traffic and influence AI summaries, so finding the “white space” took human thinking. I spent a couple hours reading online sources and searching LinkedIn for expert perspectives to get up to speed. Expert interviews weren’t available for this assignment, although the investment often adds rich insights. Then I wrote several headlines, subtitles, and a thesis myself and tested them with the editor.
3. Give AI a specific research brief
I used a paid AI research tool and gave it a detailed, multipage brief covering audience, format, angle, house style, and other inputs. I also supplied about 20 sources I had already read and a sample explainer in the target voice and format to provide context about what I was looking for in the research results.
The first results were too academic and jargon-heavy, so I pushed the AI to speak in plain English for a nontechnical business audience, as any editor would. To avoid results similar to existing technical content, I asked for current, authoritative, and primary research from consulting firms, technology companies, and academic studies, plus recent reporting from major technology outlets.
4. Have a human read the sources
The AI returned tens of thousands of words of research. I used its summaries to navigate that volume and refined the results with additional queries. Next time I will cap AI’s loquacious output to a manageable length.
I read dozens of the most relevant sources myself, including long articles and reports and some Reddit threads for anecdotal expert stories and user sentiment. I also followed footnotes and searched independently with several AI chatbots and search engines to learn what I didn’t yet understand. That research took me in directions my prompt hadn’t anticipated.
Some of the AI sources sat behind paywalls I couldn’t access. I replaced them with reporting from trusted outlets I subscribed to or that were available without a paywall.
5. Have a human build the outline
I built a detailed outline by hand. In my experience, creating the structure myself makes me absorb the research more deeply and use the information more creatively than skimming an AI-built outline would, a pattern psychologists call the generation effect. The client’s editor then checked structure and content and flagged anything missing or off-target.
6. Let AI organize the research
I used AI to sort the research into the sections of the outline where it best fit. That saved hours of organizing work and freed me to decide which data, analysis, advice, quotes, and examples would give the article more substance than a typical explainer.
7. Have a human write the draft
I drafted the article myself. I won’t detail every method for this step, because much of it is tacit knowledge accumulated over decades of experience. It doesn’t easily reduce to the automated “workflow” that AI developers unfamiliar with the process often ask to help create when I publish about AI-assisted writing.
I sometimes asked the AI to develop suggestions for charts and graphs as a starting point for iteration based on the content and to assemble information I’d gathered into well-organized tables that I heavily rewrote. AI was also helpful for suggesting more ideas for headlines, subtitles, summaries, and other routine copy than I could possibly use. Even though detectors can be inaccurate, several I tested flagged the article as nearly 100% human-written.
Human drafting is an important legal consideration for authors and organizations. US copyright law requires human authorship, and the Copyright Office has said that prompts alone don’t meet that standard. Many of the external publications where authors hope to publish prohibit AI-generated writing, as well. And since August 2026, Article 50(4) of the EU AI Act has required organizations that publish AI-generated text meant to inform the public on matters of public interest to disclose that the text was artificially generated, unless it has undergone human review or editorial control and a person or organization holds editorial responsibility for the publication.
8. Encourage AI to critique the draft
I used several AI chatbots as editors to find what a skeptical reader would question, including logical gaps, unsupported claims, weak transitions, repetition, clichés, fluff, passive voice, and factual errors. I supplied a custom list of writing patterns readers now associate with AI-generated text, although writers used many of them long before AI. I had the chatbots check the draft against the client’s formatting instructions and the SERP analysis to see where it differed from top-ranking articles.
I instructed the chatbots to deliver suggestions as tracked changes with explanatory and developmental comments, the format I’ve long used as a developmental and line editor. Then I reviewed every change and comment myself, accepting the best, researching and fixing gaps and errors, and often offering better wording than AI suggested. When I used a phrasing or idea that AI generated, I rewrote it in my own voice and fact-checked it against reputable primary sources.
9. Rewrite with human editors
AI wasn’t my only editor. The client’s editor gave feedback ranging from minor to substantial. Each additional round meant more research, pressure-testing, and rewriting, so I repeated steps 4, 7, and 8 above. I sometimes did step 8 several times until I had smoothed out the rough edges.
What AI Added to the Process
AI didn’t replace my research, outlining, drafting, or rewriting. And it honestly saved only a few hours, mostly in organizing research and generating tables, summaries, and display copy. It surfaced far more research than I could have found in the time available and gave more feedback than most editors have time to flag. That added to my workload and made me think harder as I wrote. Sometimes it was wrong: it tried to flatten my voice; flagged writing problems that weren’t problems, especially what it called clichés; and served up a few subtle factual and logical errors I had to catch. It also caught facts and logic I hadn’t gotten quite right.
I led the research, outlined the article, and wrote and rewrote it. AI served as a research assistant, logic tester, and sparring partner. I think that is a better use for AI in writing than increasing speed and output. The volume and quality of the feedback tightened the logic and the prose, with only one or two rounds of progressively lighter rewriting and editing needed.
Questions to Ask a Writer or Agency
When evaluating a writer or agency that uses AI in thought leadership, smart authors and organizations will ask:
- Who does the writing?
- Where does AI enter the process?
- How are the facts verified?
- Who stands behind the content?
- How is AI is transparently labeled and disclosed?
How Insight Content Lab Can Help
Every author and team will build its own version of this workflow to answer these questions. This is the one I’m testing. If you want a ghostwriter who works this way, or help designing and managing an AI-assisted editorial process for thought-leadership writing with clear human accountability, email me at mickey@insightcontentlab.com or visit the Insight Content Lab blog for more insights.
Frequently Asked Questions
Should AI write thought leadership articles?
AI can certainly generate reams of undifferentiated and sometimes inaccurate text, but thought leadership depends on a point of view, verified facts, and an accountable author. In my process, AI supports research and critique while a person writes every draft, which keeps the thinking original, the writing distinct, and the authorship clear.
Is AI-assisted writing eligible for copyright?
In the United States, copyright requires human authorship. The Copyright Office has said AI can serve as an assistant, but prompts alone don’t give an author rights to AI-generated writing. A human-written draft keeps authorship clear. Counsel can advise in an author’s or organization’s specific situation.
Do I need to label AI-assisted articles under the EU AI Act?
Article 50(4) has applied since August 2, 2026 to content used inside the EU. It requires “deployers” to label AI-generated text published to inform the public on matters of public interest, unless a human has reviewed it and someone holds editorial responsibility. Whether your content is in scope is a legal question to answer.
Does AI save time on thought leadership?
It can, but it often creates more work. In my experience it saved a few hours, mainly in organizing research and producing things like summaries and display copy to check and heavily revise. Its bigger payoff was depth and breadth: a wider range of research to review and more feedback on each draft than a human could normally give.