Most advice about AI and content is about speed. The evidence I found this month points somewhere else. The time did not disappear. It moved to the part where someone checks the work.
Marketers are checking what AI writes
Ahrefs surveyed 301 marketing and SEO professionals in August 2026, and Relevant Audience summarized the results. Of those surveyed, 93% said they use AI to create or help create content. Of the AI users, 89% manually check what it produces. Only 2.7% trust AI content with no review, and 66% trust it only after a human has reviewed it.
Two more interesting numbers. 78% of AI users do not tell website visitors they used AI. And 41% say AI has not saved them any money on writers or freelancers.
Put together, the tool is everywhere, the checking is nearly universal, and a large share of people are not seeing savings. Keep in mind that Ahrefs sells SEO tools and the answers are self-reported, so treat this as a rough guide, not a verdict.
One writer pointed AI at his archive
Jason Tan writes a Substack newsletter every Sunday from 8 AM. A finished article takes about two hours, and he does not use AI to draft it.
He built a Google Apps Script with Gemini Flash that reads his Substack sitemap, keeps the post URLs, and writes a one-sentence summary of each post into a Google Doc index. The problem was a familiar one. He would finish an article and realize he had missed half a dozen natural places to link to his older posts.
His first attempt used NotebookLM. It found matches but broke the formatting. In his words, rebuilding the headers and spacing took longer than doing it by hand.
Notice what the tool does here. It does not produce text for him to audit. It makes the text he already wrote easier to find, so the review step shrinks because there is less to review.
A teacher can build a simulation. Who checks it?
In May 2026, the Bangkok Metropolitan Administration and DEPA started Coding Thailand 2026 with 58 selected participants. The stated aim is to move educators from traditional roles to digital innovation coaches. A paper from Ateneo de Manila University, presented at HCII 2026, is titled "Teachers as Vibe-Coders: Authoring Adaptive Learning Simulators Through Natural Language." I only reviewed the title and venue, not the abstract.
On the practitioner side, one guide for educators estimates that differentiated assessments drop from 3-4 hours to 10 minutes plus 20 minutes of review. It walks through a sixth-grade water cycle lesson with a temperature slider and an embedded quiz. The time figure is the author's own estimate, and the guide is from April 2026. The page does not discuss how a teacher confirms the generated content is correct.
The institutions measure who attended. The guide measures how fast it goes. Neither describes the step in between, and that step matters more here than in a static handout. A typo on a handout is just a typo. The same mistake in an interactive tool is something students learn, with a slider.
When you are the only reviewer
Kyle Poyar's account of his first year solo is useful here. His newsletter reaches 88,000+ subscribers, and he calls audience size "absolutely a vanity metric." Brand partnerships are 55% of his revenue and consulting and advising 25%. He uses about 14 tools on a near-daily basis, and he writes that AI will not write his newsletter.
What went wrong was back office work. A $10k invoice took about 65 emails and 94 days to collect. He went through three business entities in a year, and multiple insurers rejected him.
A solo operator has no second pair of eyes. The person building the tool is also the person checking it and the person chasing the invoice. AI does not remove that limit. And, in my opinion it is the one that matters most.
So design the review before you build the system. If you turn a book or framework into an interactive tool, the source text is the answer key. Check every generated question, calculation, and result against it yourself, on a schedule, before anyone else sees it. The 20 minutes of review is the part worth planning.
Start with one published piece
Pick one piece of content you have already published. Decide what the checking step is and how long it takes. Then decide what AI should do around it.