Built to Be Quoted.

Answer engines do not rank pages. They lift passages.

That one sentence reorganized how I build this site. A search engine sends a visitor to your page, chrome and all. An answer engine takes 60 words of it, staples your name on, and serves it somewhere you will never see. The page is no longer the product. The passage is, and the entity behind it.

So I rebuilt this site, justinbartak.ai, to be liftable. Here is the stack, layer by layer, with the honest scoreboard at the end.

One source of truth, enforced by tests

The fastest way to lose an engine's trust is to disagree with yourself. If one page says 243,000 lines and another says 250,000, a machine deciding what to quote trusts neither.

Every citable figure on this site now lives in a single facts module: launch numbers, current numbers, career figures, and the canonical sentences that carry them. Pages import the figures; nothing restates them by hand. And unit tests fail the build when any surface drifts from the source, because a convention nobody enforces is a hope.

This is the same discipline as code: one definition, many references, and a test that screams on drift.

How does an engine know who wrote it?

Before an engine cites you, it has to resolve you. Who is this, and is this the same person as that other mention?

The site asserts exactly one Person node with a stable identifier, declared once in the layout and referenced by id everywhere else. 7 pages used to redeclare the author inline without an id, which reads to a parser as seven strangers sharing a name. Orphan declarations dissolve authority; references concentrate it.

Then the graph points outward, to identifiers I do not control: Wikidata, ORCID 0009-0005-2615-3624, and a compensation dataset deposited on Zenodo under DOI 10.5281/zenodo.19653006. The DOI is the strongest citable address the site owns: permanent, independently resolvable, and it makes me the origin of those figures rather than one more page repeating them.

Markdown twins, so machines never parse chrome

Every article here is also served as clean markdown at a parallel path, advertised with an alternate link and a canonical header back to the HTML.

Retrieval works on text. A machine reading this argument through the HTML has to strip navigation, share buttons, and an author bio to find it, and every stripping step is a chance to mangle a sentence I care about. The twin hands over the argument with nothing to strip, and a hash check keeps it from drifting from the page it mirrors.

Answer capsules, because the lift is merciless

The excerpt at the top of this post answers its own headline in under 60 words. That is not a summary convention. It is the passage engineered to survive being lifted without its page.

Published audits of cited content keep finding the same shape: self-contained definitional answers under a question heading, no links inside, one claim per capsule. Every article here leads with one, and it doubles as the summary block and the meta description, so the words a human skims and the words a machine lifts are the same words. My skeptical read of the whole AEO evidence base is its own post; the capsule pattern is one of the few findings I weight.

Does any of this actually earn citations?

Now the part most implementation posts skip.

My citation probe reads zero. Every run, published. Nothing above is proven, by my own instrument, to earn a citation yet.

llms.txt is essentially unread. A twelve-week server-log study found single-digit fetches by major AI crawlers against thousands of robots.txt requests. Mine exists, generated from the same date-gated sources as the pages, as insurance that costs nothing. Nothing lives there that is not also on an indexed page.

The biggest lever is off-site. Brand mentions correlate with AI visibility roughly three times more strongly than backlinks in the largest study I trust. The site can be perfectly liftable and still lose to being unmentioned.

So why build any of it? Because of the one investment rule that survives the uncertainty: build only what stays valuable if the retrieval landscape reshuffles next quarter. First-party data I originated, identity that resolves, structure any parser can walk, instruments that tell me the truth. All of that outlives whichever engine is winning in March. Tuning for one engine's current parser does not.

What to do Next

Pick your most-cited claim, or the one you wish were cited. Three moves, an afternoon each.

Put its figure in one place and make everything else reference it, with a check that fails on drift. Rewrite its best paragraph as a capsule: under 60 words, self-contained, answering a question someone actually asks. Then make sure the author resolves: one identity declaration with a stable id, linked to at least one registry you do not control.

Then point an instrument at it and wait. Do not skip the instrument. A site built to be quoted, without a scoreboard, is a costume.

You cannot make the engines quote you. You can make yourself the easiest correct thing to quote, and then count.

Related reading:

• AEO, GEO, and llms.txt: The New SEO the definitional playbook this post implements
• I Built the Scoreboard. It Says Zero. the instrument, and the reading
• AEO: No One Has the Answer Key the evidence base, read skeptically
• Selling to AI Agents the commercial reason machine legibility matters
• Prose Is Paint. Data Is Load-Bearing. the data infrastructure every layer above stands on
• Claude Code Is Like No One Saying No to You what unlimited compliance actually produces