A crawler used to be the audience for your site: it fetched pages, built an index, and humans arrived later. Now the first reader of your docs and your pricing page is a model that answers a question and decides whether to name you. That changes what the page has to contain. Not better prose - fewer unstated facts.
I ran the whole thing on one real site (74 URLs, developer-facing) over a single night. Here is what actually changed the picture, and what stayed out of my control.
The unit of competition is a claim, not a page
An answer engine does not rank your blog post about "the modern way to do X". It collects claims it can attribute: a number, a definition, a price, a comparison, a scope. If your site holds opinions, the model cites someone who published numbers. So step one is not "write more" - it is "find the claims you already make and make them machine-readable in the same place a human reads them".
Four artifacts that did the work
1. /llms.txt - a plain text file at the root, 60 lines, 47 links in my case: which pages to cite, in what order, with a one-line description each. It is not magic and not a standard the whole industry agreed on, but it is cheap, it is read, and it forces you to write down what your site is about without marketing adjectives.
2. One machine-readable offer. We published /offer.json as a schema.org OfferCatalog: provider, price tiers, what is in scope, deliverables, and a callToAction with a URL. Sixteen top-level keys. The point is not the format - it is that a model answering "what does this cost and what is included" now has a single unambiguous source instead of inferring from a screenshot of a pricing table.
3. Per-page structured data that matches the visible page. Service on the service page, Offer per price tier, FAQPage only where the questions are actually on the page. Markup that describes content the page does not contain is a liability, not a bonus.
4. A sitemap that is not a dump. 74 URLs, canonical tags pointing at themselves, internal links from the adjacent pages so the new page is not an orphan. Adding a page without an inbound internal link is how you publish something only you can find.
Then the plumbing: IndexNow ping after publish, and a Wayback snapshot.
Three checks you can run on your own site in ten minutes
-
curl -s yoursite/llms.txt- if it 404s, you have no stated citation policy at all. - Search your pricing page source for
"@type": "Offer"- if your prices exist only inside an image, no answer engine can quote them. - Read your top page's first 80 words. If the answer to the implied question is not there, the model is reading 4,000 words to guess.
What did not work, honestly
- Wayback save returned HTTP 500 for four consecutive runs across different pages. The existing snapshots stay live, but new coverage is not something you can force from your side. If a service is down, do not write "archived" in your notes.
- IndexNow returns 200 and that is all it returns. A 200 confirms the ping, not the index. Anyone reporting "submitted = indexed" is reporting their own optimism.
- Duplicate canonicals get refused. Publishing the same canonical URL twice through the dev.to API is blocked - which is the correct behaviour and a real constraint if you plan to repurpose one landing page across several communities. Write for a different page, or do not publish.
The part that is still a human decision
None of this makes a bad offer answerable. It makes a real offer quotable. If your service page cannot state what it costs, what it excludes, and what a buyer gets on click, AEO will just distribute the ambiguity faster.
If you want the version where someone else runs these four artifacts plus the distribution behind them, the scope and the public price list are here: https://axelfreeman.com/marketing-engineer.html?utm_source=devto&utm_medium=article&utm_campaign=aeo-service
The AEO service itself, with the artifact list we ship, is at https://axelfreeman.com/answer-engine-optimization-service.html
Top comments (0)