Rewriting a page for citation instead of ranking — a before/after teardown

The client who asked the wrong question, on purpose

A client called two weeks ago about a page that was pulling decent impressions and almost no clicks. His first question was "how do we rank it higher." I told him that was the wrong question, and he pushed back — reasonably, because for fifteen years "rank it higher" was the only question that mattered. I asked him to open ChatGPT and ask it the exact question his page answers. It gave a confident, three-sentence answer, sourced from a competitor. His page, sitting on page one of Google, never got mentioned. Ranking and citation had quietly become two different jobs, and his page was only doing one of them.

That conversation is the reason for this article. Instead of describing the difference in the abstract, I pulled a real page off our own site — the GEO tools pricing page — and rewrote a passage from it, live, to show exactly what changes and exactly what doesn't. No invented before/after traffic numbers. Just the actual text, and the actual reasoning.

Why ranking-writing and citation-writing are not the same skill

A page written to rank is written to survive a crawl and win a click. The lead paragraph often exists to set context, establish keywords, and earn trust before it gets to the point — because the reward is a click, and a click can happen after three paragraphs of buildup.

A page written to be cited is judged one passage at a time, out of context, by a system deciding whether to lift that exact paragraph into a generated answer. There is no "keep reading" credit. If the second sentence depends on the first, and the first depends on the headline above it, an engine extracting the middle sentence alone will produce a quote that reads like a non-sequitur — or it will simply skip the passage and cite a competitor's cleaner one instead.

This isn't a new idea we invented. It's the same logic behind Google's own structured data documentation, which frames schema as a way to remove ambiguity for machines reading the page — not a ranking trick. The same principle applies one level down, at the sentence level, whether or not there's a JSON-LD block attached.

The page we picked, and what it already had right

We picked our own GEO tools pricing page for this teardown because it isn't a strawman. It already ships an Article schema block, an FAQPage block, an ItemList of the tools ranked, and a "short answer" callout box up top — the exact structural checklist we'd tell a client to hit. On paper, it should be citation-ready. That's the honest version of this exercise: showing a page that did most things right, and finding the passage that still wasn't.

Here's the actual short-answer block as it reads live on the page today:

Read that paragraph completely out of context — no headline above it, no page around it, exactly the way an AI engine sees a retrieved chunk. It actually holds up well: three tools, three real prices, a direct comparison claim, no pronoun pointing at something above it. This is close to a model citable passage already. The one weakness: "covered below" is a page-relative phrase that means nothing once the paragraph is lifted out of the page — an engine quoting this sentence has no "below" to point to.

The rewrite: what actually changed

Here's the same passage, rewritten to remove the one dependency on page position, without changing a single price, tool name, or claim:

Nothing here is a bigger claim than before — it's the same three prices, the same tools, the same recommendation. What changed is that the passage no longer refers to anything outside itself. "Covered below" became an actual, self-contained sentence naming the free tools by name. That's the entire edit. It's small on purpose: the goal of a citation rewrite is never to inflate the claim, it's to remove every word that only makes sense with the rest of the page attached.

This is the same failure mode we cover in the content that gets quoted: structural patterns AI engines actually cite — page-relative language ("below," "above," "as mentioned," "see the next section") is invisible in a browser and a landmine in a retrieval system, because it's grammatically correct and semantically broken the moment the paragraph moves.

Where the FAQ block earns its keep

The same page also runs an FAQPage schema block, and the pattern holds there too. Compare the live FAQ answer to "Do small businesses even need GEO tools?":

This one was already written correctly, and it's worth pointing out why: it opens with the answer ("increasingly, yes"), makes a specific claim in the middle, and closes with an instruction that doesn't depend on anything above it. It's a useful contrast case — proof that the fix isn't "add more words," it's "front-load the answer and cut anything relative." FAQ pages versus FAQ schema for AI goes deeper on why this specific block type tends to get lifted more than prose paragraphs — it's already pre-segmented into question/answer pairs, which is most of the extraction work done for the engine in advance.

What we deliberately did not touch

Per our own house rule, nothing about the page's offers, pricing mentions, or branding changed in this exercise — this was a passage-level rewrite exercise for the article, not a live edit of that page's body copy. The prices quoted for Otterly.ai, Rankscale, and Geoptie are the same figures already cited on the page, sourced the same way. If you want the fuller structural checklist we run before touching any client page this way, the anatomy of an AI-citable page covers the six checks in order, and how to audit whether your schema is doing anything at all covers the machine-readable half of the same problem.

Two other real-world constraints worth naming honestly, because guides that skip them aren't being straight with you. First: AI crawlers have to be able to fetch the page at all before any of this matters. OpenAI's GPTBot, Anthropic's ClaudeBot, and Perplexity's PerplexityBot each publish and (per their own documentation) respect their own robots.txt directives — a beautifully rewritten passage behind a stale Disallow rule or a client-side-only render is invisible to all of them regardless of how well it reads. Second: citation is not fully controllable by either party. No honest rewrite guarantees a specific engine will choose to quote a specific passage on a specific day — what a rewrite like this does is remove the structural reasons a good passage gets passed over.

The checklist, if you're doing this on your own page today

1. Copy one paragraph in isolation. Paste it into a blank document with no headline, no surrounding page. Read it as if it arrived with zero context — because that's exactly how a retrieval system sees it.
2. Flag every page-relative word. "Below," "above," "this," "that," "as mentioned," "in this article" — each one is a dependency an extracted passage can't resolve.
3. Move the answer to sentence one. If the direct answer to the implied question is buried in sentence three, an engine has to work to find it, and a competitor's page that states it in sentence one wins the extraction.
4. Keep every number and claim identical. A citation rewrite is a structural edit, not a rewrite of your facts, your offers, or your pricing. If the numbers change, that's a different project with a different approval process.
5. Re-check readability for an actual human. A passage chopped into disconnected fragments purely for extraction reads badly to a real visitor, and real-visitor engagement is still a signal that feeds classic ranking. The goal is a passage that works read alone and read in flow — not one optimized to the exclusion of the other.

Italo Campilii

Founder, Acromatico · runs AI visibility & brand systems across 10+ live brands