Press unstick on a goal and the response includes a line from your own notes, quoted exactly.
It's a small thing and it does more work than the suggestions do.
The problem it solves
Language models produce competent, plausible, general advice, and general advice about stalled goals sounds a great deal like insight.
Break it into smaller steps. Identify what's really holding you back. Schedule a specific time. All true, all applicable to everyone, and none of it derived from anything about you. Read on a Tuesday evening when you want to feel like you've addressed something, it's convincing.
The failure isn't that the advice is wrong. It's that you can't tell, from reading it, whether the model engaged with your situation or generated the thing it would generate for anybody. Both outputs read identically.
A quoted line settles that in two seconds. Either it found something specific in what you wrote, or it didn't.
What the quote reveals
Three things, depending on what comes back.
It found the relevant line. Usually the one you'd half-noticed and not acted on — decided it should handle recurring appointments first, written in August, three weeks after the thing already worked. Seeing it pulled out and set against the advice is often more useful than the advice.
It quoted something trivial. A note about a minor session, chosen because it was there. That tells you the response is generic dressed up as specific, and you can discount it accordingly.
It couldn't find one. Which usually means there isn't much in the record — and that's a finding too. A goal with three notes in four months isn't going to produce a useful diagnosis from anything, and what it needs is contact rather than analysis.
Why verification matters more here than elsewhere
Because the failure mode of AI advice on personal goals is particularly comfortable.
Asking a model about something you're avoiding is itself a safe activity. It's engagement with the goal that carries no risk, produces output, and feels like progress. For a goal stalled on fear, that's another form of the sideways work — an evening spent thinking about the problem rather than an evening spent inside it.
Generic advice makes that worse, because it's satisfying to read and impossible to act on precisely. Specific advice, quoting your own words back at you, is harder to sit comfortably with. That discomfort is the point.
What it does to trust
Something slightly unexpected, which is that it makes the whole response easier to disbelieve.
Most product AI is designed to seem authoritative. Confident phrasing, no hedging, no visible working. That's pleasant and it makes calibration impossible — you can't tell a good answer from a fluent one.
A quoted line is a handle. When the quote is apt, you take the suggestions more seriously. When it's thin, you take them less seriously. Either way you're doing something better than accepting or rejecting the whole thing on the basis of tone.
Giving users a way to distrust your own feature is an odd design choice and it's the right one for a tool that's occasionally going to be wrong.
The alternative is worse in a way that's easy to miss. A feature nobody can evaluate gets trusted uniformly — completely at first, and then not at all after the first obviously bad answer. A feature with a visible check gets trusted in proportion, which is the only sustainable relationship with something that's right most of the time.
Where it falls short
The honest limits.
A quote proves the model read your notes. It doesn't prove the advice is good, and a well-chosen line attached to mediocre suggestions is still mediocre suggestions.
It also can't verify the interpretation. The model might quote your line correctly and draw entirely the wrong conclusion from it — decide that a note about tiredness means you need a better schedule when the actual situation is that you're dreading something.
And it's only as good as the record. If your notes are three words each, or missing for the months that mattered, the quote will be from whatever exists rather than from what would have been relevant. The technique rewards people who've been writing honestly, which is a fair distribution and not a universal one.
Twenty-two days and one line
Ship the beta to ten testers, 22 days quiet. Four notes in the previous month: refactored the db layer, redesigned settings, read about a different framework, decided it should handle recurring appointments first.
The response comes back with three moves — send it to one person rather than ten, drop recurring appointments from scope and write that down, book twenty minutes with the receptionist on Thursday. And it quotes the fourth note, observing that the requirement was invented after the tool already worked.
None of that is clever. It's the obvious reading of four notes by something with no stake in the outcome, which is precisely what he couldn't do, because he'd written each entry believing it at the time.
The quote is what makes it land. Advice saying you may be adding scope to avoid shipping would have been dismissed as generic, correctly, because it is. The same advice with his own sentence attached, dated August, is much harder to put down.
The general version
None of this requires a particular app. If you're asking any model about a stalled goal, ask it to quote a line from what you've written.
You'll get better answers, because the constraint forces engagement with the specifics. And you'll be able to tell when you haven't, which is the more valuable half.
It also quietly raises the value of writing notes at all. A record kept honestly is what makes any of this work — the quote has to come from somewhere, and a month of worked on it produces nothing worth quoting back.
Related: AI can't want it for you · Three moves of thirty minutes · Notes as evidence
Share: