25 Ways My Notes App Failed Me: A Stress Test for the Second Brain

Every notes app on the market makes the same quiet promise: type anything you remember, and we'll find it. Not keywords you happened to use. Anything you remember. It's the entire pitch behind the second-brain movement — offload your memory into an app, trust the machine to retrieve it later.

So we decided to find out if that promise survives contact with reality. We ran 25 recall tests across four popular apps (Notion, Apple Notes, Obsidian, and Evernote), using real things we'd actually written down months or years earlier. No staged data. No perfect tags. Just the messy digital attic most of us are carrying around.

The Rules of the Game

Each test worked the same way: one genuine question, one memory of the content — never the exact wording. Then we gave ourselves five minutes per app to find it using whatever tools the app offered: search, tags, AI assistants, folder archaeology, whatever.

The questions ranged from mundane to mildly urgent:

• What was the name of the plumber who quoted us $1,400 in October 2023?
• That pad thai recipe someone shared — which note is it in?
• What did we decide in the March meeting about the Q2 budget?
• Did the pediatrician say six months or nine months for the follow-up?

The last one matters. Because somewhere between the productivity-hacker fantasy and the actual stakes of daily life, there's a gap — and that gap is where these apps either earn your trust or lose it.

The Scoreboard

Across 25 tests and four apps, plain keyword search rescued us about 60% of the time. That sounds respectable until you notice which 60%. Search wins when you remember a distinctive word — a drug name, a street address, a restaurant. It fails spectacularly when you remember meaning but not vocabulary.

The pad thai recipe was the perfect example. We searched "noodles," "Thai," "peanuts," "tamarind." Nothing. The note was titled "Dinner thing??" because, apparently, past-us was very tired. Every app whiffed on that one until we scrolled manually — 14 months of notes deep.

The AI assistants fared better on fuzzy questions, roughly 70% on the ones we tested, but they introduced their own failure mode: confident wrongness. Notion's AI once cheerfully summarized a budget decision from a note that was actually a draft of the decision, later overruled. It didn't know the difference. Neither did we, for a blissful ten seconds.

Where Search Actually Breaks

Three patterns emerged, and they're worth knowing regardless of which app you use.

The title lie. Apps weight note titles heavily, and titles are the least reliable metadata in existence. "Meeting notes" appears in roughly one in five of our notes. "Ideas" is even worse. If your retrieval strategy depends on past-you writing good titles, you've built a system on vibes.

The synonym wall. You wrote "lawnmower" in June; you're searching for "mower" in November. Semantic search and AI embeddings were supposed to fix this, and sometimes they do — but only in apps where the AI index actually covers your full archive, which often requires a paid tier and a sync wait.

The tag decay curve. Tags worked great for about eight weeks after we created them. Then our vocabulary drifted — "#expenses" became "#receipts" became "#money-stuff" — and the tag system quietly splintered into dialects. If you've ever had both "#clients" and "#client" in your system, you know.

What Actually Rescued Us

When we audited the successful retrievals, the winner wasn't search or tags or AI. It was recency plus location. The notes we found fastest were the ones we could narrow by date — "sometime in October" — and then scan. Every app's list view, with dates visible, did more heavy lifting than any fancy retrieval feature.

The second-best rescue tool, embarrassingly, was the app's own link graph. Notes we'd linked to other notes (in Obsidian especially) were findable by traversal — start from the meeting you remember, follow the link to the decision you forgot. It's slow, analog-feeling, and it worked more reliably than semantic search on our messiest queries.

Trust, But Verify

Here's the uncomfortable takeaway: the second-brain pitch assumes retrieval is a solved problem. It isn't. Search is a guess about how your past self talked, and AI is a guess about what your past self meant. Both guesses fail most often when the stakes are real — the medical detail, the financial figure, the thing your boss asked about in a meeting you barely remember.

Our practical advice after 100 total lookups:

• Write for future search, not present beauty. Put the distinctive noun in the first two lines, not buried in a paragraph.
• Date everything, even loosely. Recency filtering rescued more queries than any smart feature.
• When it matters, verify against the source. AI summaries are excellent starting points and terrible endpoints.
• Keep a tiny "hot notes" list for the 5% of information you actually need repeatedly — your second brain should have a front pocket.

None of this means ditch your app. It means demoting it from oracle to library. A library with a mediocre card catalog is still useful — as long as you don't mistake it for a memory.