7 AEO Mistakes That Cost You AI Citations
The AEO mistakes that cost the most citations are burying your answer, stuffing keywords, and blocking the crawlers that feed AI tools. In the research that named GEO, keyword stuffing scored about 8% below doing nothing, while adding quotes and statistics lifted visibility 30% to 40%. Most fixes take minutes.
Short answer: answer in the first sentence, back it with a number, and make sure AI crawlers can reach the page. Those three fixes clear four of the seven mistakes below.
1. Burying the Answer
Burying the answer is the most expensive AEO mistake, since engines quote the page that answers in its first sentence. Microsoft’s Bing team says it bluntly in its guidance for AI answers: “Don’t hide important answers in tabs or expandable menus” and “Avoid long walls of text.”
So put the direct answer right under the heading, then explain. If a reader saw only that one sentence, they should still walk away answered.
Take a returns page as the example. The buried version opens with two paragraphs about how much the brand cares, then mentions the window halfway down. The version that gets quoted opens with “You can return any unworn item within 30 days for a full refund,” and the story comes after.
That habit started with the snippet box, where Google lifted one short passage to the top of the page. The surfaces have multiplied since then, which is why AEO is not just featured snippets anymore, but the passage that wins is still the one that answers first.
Writers trained on suspense fight this one. But suspense is for novels, and nobody quotes a novel to answer a shipping question. To test your own site, read only the first sentence under each heading and ask whether a stranger would walk away answered.
2. Treating AEO as a Separate Project
Treating AEO as a separate project from SEO doubles your cost without doubling the work. Google’s page on AI features says “there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.”
| What the Add-On Bills For | What It Usually Is |
|---|---|
| AEO audit | A content audit with one extra column |
| Answer optimization | Rewriting openers, which good SEO copy already does |
| llms.txt setup | A file Google says it ignores |
| AI visibility report | Search Console plus a monthly prompt log |
The overlap between the two is nearly total: crawlable pages, direct answers, proof. A page that can’t rank or earn a snippet rarely gets quoted either, so the fundamentals still come first.
What does change with AEO is small. You format answers so they can be lifted, you keep your product feed clean for shopping results, and you add AI citations to the monthly report. All three fit inside work an SEO retainer already covers.
That’s why, if a proposal charges you for both on the same pages, you should ask what the second fee buys before you hire an AEO agency. The money usually does more good making one more page worth quoting.
3. Stuffing Keywords Into the Answer
Stuffing keywords into your answer lowers your odds of being quoted, and this one has receipts. In the GEO paper’s tests, keyword stuffing scored 17.7 on the main visibility measure against 19.3 for pages left alone, and it did 10% worse than the baseline in the follow-up test on Perplexity.
Answer engines are language models, and stuffing reads as spam to a system built on language. Write the answer the way you’d say it across a counter, with the keyword appearing once because it belongs there.
Here is the difference on a product page. The stuffed version reads “Our heated jacket for women is the best women’s heated jacket for women who need a heated jacket.” The natural version reads “This heated jacket runs 10 hours on low and fits true to size.”
The second sentence uses the phrase once and gives the engine two facts to quote. Read the answer out loud, and cut every phrase you would never say to a customer.
One more thing about that paper: the edits that won were all about adding substance, and none were about repeating words. I make the longer case in writing for AI or for people.
4. Publishing Claims Without Proof
Publishing claims without proof leaves the biggest tested gain on the table. The same paper found that adding quotations, statistics, and cited sources produced the top results, with the best methods improving visibility 30% to 40% over the baseline.
The effect was largest for the underdogs. For sites ranked fifth in Google, citing sources raised visibility by 115.1%, while the top-ranked site lost 30.3% on average.
Think about the engine’s job for a second and that makes sense. It has to pick a handful of sources to stand behind, and a page that shows its receipts is a safer bet than a page that just sounds sure of itself.
On a store, the unproven version is “our customers love how warm it is.” The proven version is “it holds a 4.8-star average across 1,200 reviews, and 9 in 10 buyers say it fits as expected.”
So give the engine something to hold. A stat from a primary source, a quote from someone credible, or a number from your own order data gives an engine a reason to pick your page over ten identical ones. Filler stats don’t count, since every competitor is pasting the same three.
Also, link the source where the claim lands instead of piling links at the bottom. That is point 6 on the 10-point AEO checklist, and it is the cheapest edit on the list.
5. Blocking AI Crawlers by Accident
Blocking AI crawlers by accident zeroes out everything else on this list, and it happens in two ways most store owners never check.
The wrong bot in robots.txt
The wrong bot in robots.txt is the first way. OpenAI’s crawler documentation lists separate bots with separate jobs: GPTBot collects training data, and OAI-SearchBot decides whether you appear in ChatGPT search answers.
To check yours, open yourstore.com/robots.txt and search for “OAI-SearchBot” and for any “Disallow: /” line sitting under a bot name. OpenAI notes that a robots.txt change takes about 24 hours to show up in search results, so a fix today is live tomorrow.
Just know that a copy-pasted rule blocking every OpenAI bot takes you out of both training and search. So decide on purpose which crawlers you allow, using the tradeoffs in should you block AI crawlers.
Content that only loads with JavaScript
Content that only loads with JavaScript is the second way. Vercel and MERJ studied crawler traffic and found that “none of the major AI crawlers currently render JavaScript”, which covers the bots behind ChatGPT, Claude, and Perplexity.
On a store, that usually means reviews, size charts, and tabbed specs an app injects after the page loads. Google can see them, but the chat tools can’t. To check a page, view its source and search for a sentence from your size chart. If the sentence is missing from the raw HTML, the crawler never read it.
6. Hedging Every Answer
Hedging every answer keeps you out of the quote box, since an engine looking for a committed answer skips the page that opens with “it depends.”
“Shipping times vary based on several factors.”
Nobody quotes that sentence. “Standard shipping takes 3 to 5 business days” gets quoted, and the exceptions belong in the paragraphs after it, where they make you look careful.
Hedging shows up most on the questions that drive sales: sizing, shipping, returns, and “which one should I buy.” Those are also the questions shoppers bring to ChatGPT, so a hedge there costs you twice.
Microsoft’s guidance makes the same point about vague wording. It warns against claims like “next-gen” with nothing behind them, since a model can’t repeat a claim it can’t pin down.
The same goes for product comparisons. “Both jackets are great options” gives an engine nothing, while “pick the lightweight one for commuting and the insulated one for days below freezing” gives it a recommendation it can repeat with your name attached.
If a question has no single answer, pick the one that is true for most of your customers and name who the exception is. That is still a commitment, and it is the core idea behind answer engine optimization. Lawyers hate this advice, which is a decent sign it works.
7. Measuring Only Clicks
Measuring only clicks makes working AEO look broken. Ahrefs measured a 58% lower click-through rate for the top-ranking page when an AI Overview is present, comparing December 2023 with December 2025 across 300,000 keywords.
So a page can get quoted more and clicked less in the same month. Score yourself on these instead:
- Impressions in the AI features report in Search Console
- Sessions and orders from AI referrers in GA4
- Branded search volume, since some shoppers read the answer and then search your name
- A monthly log of who gets cited for your top ten questions
On that last one, don’t trust a single run. SparkToro and Gumshoe had 600 volunteers run 12 prompts 2,961 times and found less than a 1 in 100 chance of getting the same brand list twice. Run each question several times and track how often you appear, not where you rank.
None of this means clicks stopped counting. The scoreboard grew, and a team staring at one column is reading the game wrong. Building that wider scoreboard is part of my AI search services for Shopify stores.
How Do You Check a Page for AEO Mistakes?
You check a page for AEO mistakes by running seven quick tests, one per mistake, which takes about 15 minutes a page. Start with the page that already gets the most impressions in Search Console.
| Mistake | The Check | A Pass Looks Like |
|---|---|---|
| Buried answer | Read the first sentence under each heading | A stranger is answered by that sentence alone |
| Separate project | Look at your invoice | One line for search, not two |
| Stuffing | Read the answer out loud | Nothing you wouldn’t say to a customer |
| No proof | Count the numbers and linked sources | At least one per section |
| Blocked crawlers | Check robots.txt and view source | OAI-SearchBot allowed, text in the raw HTML |
| Hedging | Search the page for “depends” and “varies” | Neither word opens a section |
| Clicks only | Open your monthly report | It has a citations column |
Any page that fails three or more goes to the top of your rewrite list.
Is missing schema an AEO mistake?
Missing schema is not an AEO mistake for Google’s AI features. That same Google page says “there’s also no special schema.org structured data that you need to add.”
Still, keep the product markup your Shopify theme already outputs, since regular rich results use it. Just don’t pay extra for “AI schema.”
Is a short page an AEO mistake?
A short page is not an AEO mistake when it answers the question. Engines quote passages, and a 300-word page with a clear answer and one real number beats a 3,000-word page that never commits.
The length problem runs the other way more often. Long pages bury answers, which sends you right back to mistake 1.
Which AEO Mistake Should You Fix First?
Fix blocked crawlers first, since mistake 5 is the only AEO mistake that makes every other fix worthless. It also takes the least time: ten minutes with robots.txt and one view-source check.
After that, go to mistakes 1 and 6 on your ten highest-traffic pages. Both are rewrites of a single sentence per section, and they change what an engine sees first. Proof and tracking come last because they take the longest to build.
Want a Second Set of Eyes on Your Answers?
Every mistake on this list breaks one of three rules: be reachable, answer first, or prove it. Here is what I’d do this week:
- Check robots.txt for OAI-SearchBot and view source on your bestselling product page
- Rewrite the first sentence under every heading on your top five pages
- Add one number from your own store to each of those pages
- Book a call and I’ll read those five pages the way an answer engine does, or hand the whole list to my AI search services
Sources:
- Aggarwal et al.: GEO, Generative Engine Optimization (KDD 2024)
- Google Search Central: AI Features and Your Website
- Microsoft Advertising: Optimizing Your Content for Inclusion in AI Search Answers
- OpenAI: Overview of OpenAI Crawlers
- Vercel: The Rise of the AI Crawler
- Ahrefs: AI Overviews Reduce Clicks by 58%
- SparkToro: AIs Are Highly Inconsistent When Recommending Brands
