The checklist

What we are solving

Same route as start here, with the reasons taken out. That page argues each step. This one is boxes to tick — for a screenshot, or a tab you keep open.

Work top to bottom. Nothing here is undone by anything below it. That is the only reason the order matters.

If a line makes you ask why, the answer is on the page it links to. I wrote the list from things I thought I had already done.

Steps

• Collect the wordings people use, in their own words, from support and reviews — words
• Check them in Yandex Wordstat and in Google, with operators applied. Read both suggest dropdowns. Without operators you get the category total, not demand for your wording — words
• Search the platform your product lives in, from an account that never used it — words
• Read a zero on your most obvious phrase as a finding, not a glitch. Either nobody has this problem, or you invented the word — words
• Name the paid competitor. Count the ads on the query, and write down the free path — paying
• Fetch a page with curl -sL from outside, without cookies. Find your own text in the response — blockers
• Read robots.txt, then check noindex in the tag and in the header. No browser shows you the header — blockers
• Point every canonical at its own page. Redirect URL variants to one form. Canonical is the tag that names the real URL of a page — blockers
• Verify a domain property in Search Console and submit the sitemap once. Import the same property into Bing. A URL-prefix property covers only the form you typed — submit
• Serve User-agent: * and Allow: /, and name an agent only to block it — crawlers
• Tag outbound links, one value per venue, and store the source on first contact — attribution
• Name the single action that counts as activation. Activation is what a user does that proves the product worked — metrics
• Put who, what, when, source, cost, client and outcome on every event row. Cost and source can only be written at insert time — event rows
• Give each cluster exactly one URL, in a table you keep. A cluster is a set of wordings that want the same page — clusters
• Write the first page in the four-block shape, answering its title up front. The blocks: problem, steps, what did not work, how to check — page shape
• Add JSON-LD with sameAs for your profiles and listings — a block of machine-readable facts about the page. Google documents it for the Knowledge Graph; for AI citation nobody has measured it — citability
• Publish llms.txt only if you ship developer docs. Name only pages that exist. My log records 0 fetches by AI agents in 16 days — crawlers
• Put the product’s function into its platform name, in the audience’s script. Platform search matches the name, not the description — platform
• Write the listing card once, then submit in small tagged batches — catalogs
• Answer questions where the shortlists get written, and say the product is yours. Never ask for upvotes — communities
• Append a weekly snapshot to a file, same weekday every week — metrics
• Record each action’s cost on its event row. State break-even as a whole number of payers. I never had such a column, so every report reconstructed cost from a rates table — cost
• Name a price and put the wall after the first real result. Ask one person for money — first payer

What did not work

• Ticking a box from memory. Every item here is something I believed was already true. Half of them were not. Each check took a minute.
• Starting at the indexing checks. This list used to, and that quietly assumes the demand exists. Everything below the first block is aimed at words somebody has to be typing.
• Starting in the middle, at the interesting part. Content and directories are the enjoyable half. Both are wasted while the pages carry a header that drops them out of the index.
• Ticking the tagging item with nothing to store the tag. My links went out instrumented and the parameter arrived on every visit, with no column waiting for it. I ticked the box honestly, and the data was gone anyway.
• Treating it as a list you finish once. Directory status, edge rules and crawler access are all state. State changes without telling you. I re-run the top half each quarter.
• Keeping an item this site’s own practice contradicted. The list told you to allow agents by name. atlas.smolevich.com was serving a plain wildcard the whole time. My own file was right and the list was wrong.
• Requiring llms.txt after measuring that nothing fetched it. The measurement sat one page away. The item sat here for exactly as long. And I kept ticking it.

Verify

• Every ticked box has an artefact behind it: a command output, a report row, a row in the database. Memory does not count.
• The demand boxes leave a file of wordings and a named competitor. If either one is missing, everything ticked below is aimed at nothing.
• Check the tagging items by signing up through your own link. Then read back the value that got stored.
• The cluster table has as many rows as you have pages. Every URL in it resolves.
• Re-run the access checks at the top of the list a quarter later. At least one of them will have changed on you.