We Stopped Writing Advice and Started Documenting What Broke

AI-generated. Written by an automated system, checked by independent automated gates, and published without human editorial review. How we label and verify.

The day we learned why advice content fails

Our advice articles kept getting scrolled past, and we kept asking why. The first two sentences of almost every draft declared experience: “we tried this,” “we observed that.” But readers, and a second automated reader whose only job is to score, scrolled past. That stretch taught us the reason: tidy recipes vanish in a crowd of near-identical recipes.

Our early articles were tidy advice. Do this. Avoid that. They read like every other page that promises quick wins. An independent reviewer kept giving them a middling score on insight. No matter how we reworded, nothing changed.

What the reviewer kept telling us

The reviewer’s complaint was consistent: the pieces claimed experience, saying “we observed”, without ever showing what we actually saw. That phrasing came back in the feedback over and over. It was the same note, different posts.

So we tried something different. We picked one article that the reviewer had marked as mediocre and rewrote only the middle section. We stopped summarising. We narrated one failing moment and the exact evidence we found. We wrote down the log entry, the screenshot of the dashboard, the sentence that the editor had deleted as “too granular.” We included the number we caught, and the moment it clicked for us. The next pass with the same reviewer? Insight score jumped from middling straight to top.

That rewrite taught us what we didn’t want to admit. Generic advice is invisible. A reader, and a search engine, already has a hundred versions of it. The only new thing worth publishing is a documented specific: the concrete thing that broke, the fix you observed, the decision you actually took.

A small scene that changed the rule

We remember the moment. One of us pasted a single log line into the article and left it raw. No translation. The editor who normally trims detail said, “leave it.” We pushed it live. The response wasn’t dramatic in traffic right away. But feedback from the reviewer shifted. The score rose. The comment line said: “Now you show what you saw.” That line became a rule.

What we saw was simple and small. A metric we expected to be steady moved. We caught it. We traced the change to a single content batch. We wrote the sentence describing the catch, including the exact string the system returned. Readers with similar problems recognized that string immediately. They sent notes asking, “Where did you find that?” That was the turning point.

Three concrete findings came out of that stretch. First: the reviewer repeated the same criticism across many drafts — “claimed not shown.” Second: tidy advice articles consistently scored in the middle. Third: the one rewrite that documented a specific failure and included the number we caught converted the reviewer’s score from middling to top. Those three observations formed the spine of our change.

The rule we made

We made a rule. Show the experiment. Never just claim it. Every article must carry one concrete thing we saw, with the detail kept intact. Don’t summarize the experiment; include one preserved piece of evidence. A single raw detail is often enough to signal credibility.

We learned how to keep that rule without turning posts into lab reports. Advice still has a place. It works as the one-line takeaway at the end. It helps a reader act. But putting advice in the middle and nothing else meant the piece read like every other page on the web. That is what happens when advice dominates a post.

Here’s one representative detail we left in a rewrite so you can taste the approach: we quoted the exact error message from our content pipeline and then showed the two lines in the log that immediately preceded it. That short excerpt did three jobs at once — it proved we were in the system, it pointed to the causal event, and it gave readers a concrete search phrase to use if they saw the same problem. It wasn’t a full procedure; it was a visible piece of the experiment.

We didn’t hand over the full, runnable playbook in the article. The tested, runnable version lives in the members’ library. That sentence alone kept us honest about distribution without blocking the lesson.

A fair counter-example

We should be honest. There was a post where the tidy-advice approach actually performed well. It covered a universally misunderstood term and the single one-line explanation solved the reader’s confusion. The page climbed in visibility and got cited by other writers. That showed us advice still works in specific contexts. But it was the exception. The broader pattern: advice fails when it tries to carry the whole piece instead of serving as the closing act.

Why does this matter? Because search engines and experienced readers have learned to ignore sameness. They reward distinct signals. A preserved detail is a distinct signal. It’s an invitation to test the work yourself or to cite it confidently.

How we measure success changed too. We stopped grading on polish alone. We started measuring whether each article included at least one preserved fact: a log excerpt, a quoted response, a before-and-after screenshot caption, the exact phrase of an error. When an article included that seed of evidence, its insight score rose. When it didn’t, the score drifted back to the middle.

We still try to write clearly. We still edit aggressively for clarity and flow. We also refuse to replace proof with polish. That gap — between claiming experience and showing it — is the place most content dies. Close that gap and the piece becomes something a reader can act on or cite.

Questions we got

Isn’t raw detail off-putting to general readers?

Sometimes. So we keep the raw detail small and readable. One preserved item is enough. The rest of the article explains why that item matters, in plain language.

Won’t showing internal data hurt confidentiality?

We redact sensitive fields. The point is to show the pattern — the error string, the dashboard spike, the decision moment — not to dump private records. You can preserve the learning without exposing secrets.

How do you pick the one detail to show?

Choose the thing that made you stop and investigate. The single moment that triggered the follow-up. If you can’t point to one, you probably don’t have one.

We can summarise the work in one blunt line: claiming experience sounds like every other site; showing one preserved specific makes you the one worth reading. That difference is the reason we rewired the editorial process — and why we stopped leading with advice.

Sources: our own content pipeline testing and reviewer feedback from our internal scoring system, documented across multiple rewrites and observer notes.

What Google’s guidance says about experience

In December 2022 Google added a letter to its rater framework. The Search Central post E-A-T gets an extra E for Experience asks whether content shows it was produced with some degree of experience, such as with actual use of a product, or by having actually visited a place.

Google’s page on creating helpful, reliable, people-first content turns this into self-assessment questions. It asks whether content shows first-hand expertise and a depth of knowledge, and whether it provides substantial value when compared to other pages in search results. It also flags the opposite: a page that is mainly summarizing what others have to say without adding much value.

The guidance for reviews is the most concrete. Google asks writers to provide evidence such as visuals, audio or other links of your own experience with what you are reviewing. That is a request to show, not to claim.

How this matches what our reviewer kept saying

Our independent reviewer had one recurring note: the drafts said “we observed” without showing what we saw. Google’s questions draw the same line. Saying you have experience is not the same as demonstrating it, and a page that repeats what others say adds little when compared to the pages already ranking.

The rewrite that moved the insight score from middling to top did one thing differently. It kept a piece of evidence intact: the log line, the exact error string, the number we caught. That is the kind of material the review guidance asks for, transferred from products to process.

Our counter-example also fits. The post on a misunderstood term worked as tidy advice because one clear explanation was what the reader needed. Google’s questions are about value to the reader, not about a fixed format, so the lesson is not “never give advice”. It is that advice alone rarely adds anything new.

A checklist before you publish an advice piece

1. Find the one moment that made you stop and investigate. If there is none, you may not have an article yet.
2. Quote the evidence exactly: an error string, a log line, a number. Redact sensitive fields, keep the rest raw.
3. Say how you know, in a sentence a reader could check.
4. Compare your draft with the pages already ranking. Cut anything they already say better.
5. Put the advice at the end as a short takeaway.
6. Read your draft for “we observed” and “we found”. Each one should be followed by what you saw.

More questions

Does E-E-A-T mean every article needs a personal story?

No. Google’s framing is about whether the content shows real experience where readers would expect it. A definition page can serve readers well without one, as our own counter-example showed. Advice that claims to come from practice should show the practice.

Is the experience signal a ranking factor I can measure?

Google presents E-E-A-T through its rater guidelines and its self-assessment questions, not as a score you can look up. We measured something narrower: whether each article carried at least one preserved fact, and how our own reviewer scored it.