Adding first-hand experience to AI drafts (the E-E-A-T fix)

Google's quality framework goes by E-E-A-T, and the first E is the newest — added specifically for the question 'has this author actually done the thing they're writing about?' It's also the exact dimension AI drafting can't supply. A model can produce competent expertise-shaped text about anything; it has never installed a water heater, filed a trademark, or dealt with your city's permit office in July. Which means the first E is where your content either earns its existence or joins the undifferentiated pile.

What E-E-A-T actually stands for

The four words, from Google's own search quality guidelines, each answer a different reader question:

• Experience means the author has personally done the thing the page is about. A plumber writing about water heater replacements has it. A writer who researched other articles about water heaters doesn't.
• Expertise means the author knows the subject well enough to give correct advice — the training and knowledge behind the work, not just the repetitions of it.
• Authoritativeness means other people treat you as a source: they link to your site, mention your business, review you, recommend you. You can't write this one into a page — it's earned outside it.
• Trustworthiness means a reader can safely act on the page: the facts are accurate, the prices are real, nothing is exaggerated or hidden. Google treats this as the most important of the four.

AI drafting can imitate the shape of expertise and can't touch the other three — they come from who you are and what you've done. That's not a reason to avoid AI drafting. It's a reason to treat the draft as the container and your experience as the contents — and to have a repeatable way of filling it.

What experience looks like on the page

Raters and readers can't verify your résumé; they can only read the page. Experience shows up as textual evidence — details that would be hard to write without having been there:

• Your numbers — what jobs cost, how long they took, how many you've done, what failed and how often. First-party data is the strongest experience signal there is, and the same specifics AI search engines quote.
• Cases — 'last winter a client called with X; here's what we found and what it cost to fix.' Anonymized is fine; specific is mandatory.
• Surprises — what beginners get wrong, what the standard advice misses, what you changed your mind about. Models can't be surprised; people who've done the work always have a list.
• Process artifacts — your actual checklist, the questions you ask on the first call, photos from real jobs. Things that exist because the work happened.
• Judgment calls — 'we stopped offering X because…', 'we recommend against Y for houses older than…'. Opinions with reasons attached are experience wearing its work clothes.

The workflow: interview yourself before the model writes

The common failure is asking AI to draft, then trying to sprinkle experience onto finished prose — it doesn't stick, because the structure never made room for it. Reverse the order. Before drafting, spend ten minutes answering five questions in rough notes: What do customers always ask about this? What does it actually cost or take, in your recent jobs? What went wrong recently? What do people believe about this that's wrong? What would you tell a friend off the record? Feed the notes into the prompt with the instruction to build the post around them — the model's job is organizing your material, not sourcing its own. (That instruction also starves hallucinations, which cluster wherever the model has to invent specifics you didn't supply.)

Don't add experience to the draft. Draft around the experience. Ten minutes of notes before prompting beats an hour of decorating afterward.

A paragraph-level test

Editing an AI draft, run each section past one question: could a competitor who has never done this work publish this paragraph unchanged? If yes, the paragraph is packaging — either inject something from your notes or cut it. A finished post doesn't need every paragraph to pass; it needs enough failures-of-the-test (in your favor) that a reader finishes knowing they read someone who does the work, not someone who researched the phrase. Three genuinely experience-bearing passages will carry a post; zero will sink it — that's the sameness problem in its purest form.

Sign it like you mean it

Experience evidence works harder with a person attached. A real author name with a two-line bio stating what they do and how long they've done it; 'by the team' pages reserved for genuinely collective work; photos that are actually yours. This isn't ceremony — anonymous expertise-shaped content is the default output of the entire internet now, and the sites that got demoted for it were overwhelmingly the anonymous kind. Attribution is cheap differentiation.

Why this is the durable strategy

Every capability of AI writing — fluency, structure, coverage — is available to your competitors at the same monthly price, and it shows: the median blog post in every niche is converging on the same competent emptiness. First-hand experience is the input that doesn't commoditize, because it's generated by running your business, which competitors can't subscribe to. The practical conclusion is almost cheerful: the work you're already doing produces a continuous stream of the one ingredient content marketing can't fake. The workflow above is just a pipe from the job site to the page.