Every Tuesday, we take one brand apart in public.
Twenty buyer questions. Four answer engines. One brand, and the honest reason the assistants name it or walk past it. No theory, no predictions about where search is heading — just what four machines actually said this week, and what we would change on Monday.
Quoted, and still not recommended
There is a particular kind of losing that looks like winning. Your page gets cited. The assistant quotes your research, borrows your framing, uses your numbers — and then recommends somebody else. Ahrefs measured it across 9,886 answers: 43% of the ones citing their own page never named their brand at all.
This week we pull apart a brand doing exactly that. Three pages that get quoted constantly, the competitor being recommended straight out of them, and the small structural habit causing it — a claim and a category sitting one paragraph apart, which is all the distance a model needs to take one from you and the other from someone else. The rewrite is in there, before and after, with the ratio re-measured a fortnight on.
- Why the distance between your category term and your differentiator decides who gets named
- The competitor-adjacency audit: whose name is sitting next to yours on the pages you already win
- A two-number scoreboard, so a citation that earned nothing shows up as the loss it is
You are paying to be found and blocking the door
The most common finding in every audit we run, and the one that takes ten minutes to fix. Somewhere between your CDN, your security plugin and a default nobody chose, GPTBot and PerplexityBot are being turned away at the door — usually filed alongside the scrapers, usually by a checkbox somebody ticked in 2023 to stop content theft.
Nothing downstream of that matters. Not the schema, not the rewrites, not the outreach. The page is never read. So this issue is the ten-minute check, done four ways — Cloudflare, Hostinger, Vercel, AWS — with what each provider’s defaults quietly do, and how to let the retrieval crawlers in while keeping the training ones out, if that is where you want the line.
- The exact robots.txt block to paste, and why Google-Extended does almost nothing people think it does
- Reading your own logs to prove the agents are arriving — permission and arrival are not the same event
- Training versus retrieval: where the distinction actually holds, and how to opt out of one without losing the other
Read it on Tuesdays
One email a week. The teardown, the numbers under it, and nothing else — no drip sequence waiting behind the signup, no webinar invitation, no second email on Thursday. Leave in one click. The list is never rented, sold or shared, and 11,400 (as of 14 November 2026, per syed.ai) people have found that worth their inbox.
What subscribers get
Cadence — Tuesdays, 52 a year
Format — One teardown, ~1,200 words
Prompts run — 20 per brand, 4 engines
Subscribers — 11,400 (as of 14 November 2026, per syed.ai)
Cost — Free, no upsell tier
Put your own site in the box above and it joins the queue. If it gets picked, you read the findings the same morning everyone else does — no preview, no right of reply. That is the deal, and it is the only reason the teardowns are worth reading.
Want us to look at your site?
Put your site in the contact form and say it is for a teardown. If it is picked, you read the findings first.