The expensive AEO mistakes we keep seeing
The five we meet most: a site AI crawlers cannot read (blocked or JavaScript-only); optimizing for one prompt when assistants expand every question into many; chasing word count when freshness and format drive citations; treating a lucky screenshot as proof; and doing all the work on your own domain when off-site mentions are the strongest lever. Each one burns months. All five are avoidable in the first week.
What are the most common AEO mistakes?
Mistake one: being unreadable and not knowing it
The most expensive failures are invisible: a robots.txt line from an old template, a CDN blocking AI bots by default, or a JavaScript-only site serving crawlers an empty shell. Everything downstream — content, mentions, ads pointing at the site — underperforms while the cause sits in plain sight. The five-minute check with real bot user agents and JavaScript disabled comes before any strategy conversation, every time.
Mistake two: optimizing prompts instead of topics
"Get us into ChatGPT's answer for this exact question" misunderstands the machine. Assistants expand each prompt into around nine to eleven synthetic subqueries, over 95% of which have no recorded search volume, and people phrase the same intent endlessly. Coverage of the topic — one strong page per real question, interlinked, plus mentions across the category — is what makes you show up regardless of phrasing. Single-prompt optimization is astrology with a dashboard.
Mistake three: the word-count reflex
The instinct that more words equal more authority dies hard, and it is measurably wrong here: 0.04 correlation between length and citation, 53% of cited pages under 1,000 words. The budget wasted on 4,000-word pillar pages is budget not spent on the monthly refresh cadence that actually correlates with staying cited. Depth belongs where a question genuinely needs it; everywhere else, current and precise beats long.
Mistakes four and five: screenshot proof, and homepage myopia
AI answers vary run to run — Ahrefs measured about 45% of citations changing between refreshes. One screenshot of ChatGPT naming you proves nothing and, worse, one screenshot of a competitor being named proves nothing either. Real measurement is repeated runs, multiple platforms, tracked monthly. Any vendor reporting wins by screenshot is selling weather as climate.
And the biggest allocation error: spending everything on your own site. Branded mentions across the web correlate with AI visibility at 0.664 — stronger than any on-site factor — and YouTube presence correlates with ChatGPT visibility at 0.737. If your plan has no off-site motion — no directory strategy, no lists, no communities, no video — it is at most half a plan.
We published thirty AI-optimized pages and nothing happened. Why?
Usual suspects, in order: crawlers cannot read them (check first, always), the pages answer no real question, nothing off-site mentions you, and it has been six weeks in a game whose organic side moves in months. The audit sequence exists precisely to find which of these applies before more money follows.
Is it a mistake to start AEO before our SEO is fixed?
They are not sequential. Crawler access and answer-first content help both immediately, and mention-building compounds regardless of your rankings. What is a mistake is buying AEO retainers while your site is unreadable to AI crawlers — fix the floor first; it takes days, not quarters.
Should we respond when ChatGPT says something wrong about us?
Yes, at the source. Assistants repeat what the web says, so find the stale or wrong pages feeding the claim — often your own — and correct them, then publish specific official facts (pricing, service area, what you actually do) so there is no vacuum. Wrong beliefs propagate; specific truth displaces them.
What is the single highest-leverage first move?
The technical check, because it is fast and everything depends on it. Second: find the three lists or directories assistants cite most in your category where competitors appear and you do not. Getting onto cited pages moves the needle faster than anything you publish yourself.
Last reviewed 23 August 2026