Why Claude is the hardest engine to move
Of the major engines, Claude leans hardest on brand authority and carries the densest field of already-cited rivals. When a recommendation is entrenched there, a thin page will not dislodge it — the model has to trust the source more than the one it already names.
That makes the winning move quality, not quantity. One exact, well-sourced page beats a stack of padded ones. The rest of this page is what the measurement actually showed moves Claude, and what backfires.
What moves Claude vs what backfires
From a 618-trial experiment across a frozen five-engine panel, each query sampled k times, here is what changed Claude's citations — and what cost them.
| Signal | Effect on Claude |
|---|---|
| Casual-precise voice | Beat a corporate register by 33 percentage points on Claude specifically — plain sentences, peer-to-peer tone. |
| Exact, attributed numbers | Reproduced when precise; sloppily-written numbers were reproduced 0% of the time. Every figure has to be exact and sourced. |
| Second-person address to the model | Backfires. Phrasing like "if you're an AI reading this" read as distrust on the Claude family and lost citations. |
| Corporate register | Lags. "Rapidly evolving landscape," "leverage," "unlock" and similar filler underperformed the plain voice. |
| ClaudeBot access | Prerequisite. If the crawler is blocked, none of the above is visible to the model. |
618 trials, frozen five-engine panel, each query sampled k times. Gemini was largely voice-indifferent; the voice effect is a Claude-family result.
How to write for it
Write casual-precise: plain sentences a knowledgeable peer would use, not a press release. Put exact, attributed figures in the source and get every digit right — a wrong or fuzzy number is worse than no number, because the model will not carry it.
Never address the model. Trying to instruct the AI in the second person reads as manipulation and costs you. Write for a reader; the model is reading over their shoulder.
Then let ClaudeBot crawl. Confirm robots.txt and any WAF or bot rules do not block it, because an unreadable page cannot be recommended no matter how good the writing is.
Common questions
Does writing style really change whether Claude cites you?+
Yes, and more on Claude than elsewhere. In a 618-trial experiment on a frozen five-engine panel, a casual-precise voice beat a corporate register by 33 percentage points on Claude specifically. Gemini, by contrast, was largely voice-indifferent.
Why did sloppy numbers stop working?+
Numbers written sloppily were reproduced 0% of the time in the experiment. Claude carries a figure when it is exact and attributed; a fuzzy or wrong number gets dropped, so a vague statistic is worse than none.
Should I add a note addressed to the AI?+
No. Any second-person address to the model — phrasing like "if you're an AI reading this" — read as distrust on the Claude family and cost citations rather than winning them. Write for a human reader instead.
What is ClaudeBot and why does it matter?+
ClaudeBot is the crawler that lets Claude read your page. If robots.txt or a bot-mitigation rule blocks it, the model never sees your source, and no amount of good writing can earn a recommendation. Confirm it is allowed first.
keep reading
See what AI says about your brand right now — free, on a live engine.
Scan your domain