Not gonna lie, I'm suspicious of most AEO checklists. They're forty items of "add schema" with zero evidence and no sense of what anything costs you in hours. So this list plays by different rules: 17 tweaks, grouped into one realistic weekend, each with an effort estimate and, wherever it exists, the actual data. Where the evidence is thin, I'll say so. The stakes are real either way: AI Overviews now show up on nearly 60% of Google queries, and the average one now cites 15 sources, double what it linked a year and a half ago. More seats at the table, still mostly unclaimed.
Friday Night: The Plumbing (an hour, tops)
1. Unblock the AI Crawlers
Start here because nothing else matters if the bots can't get in. Ziptie's crawler research found 27% of sites block at least one major AI crawler, usually by accident: an old robots.txt rule, a security plugin, an overzealous WAF, or Cloudflare, which has blocked AI crawlers by default on new domains since July 2025. Know which bots matter, too. GPTBot trains models; OAI-SearchBot is what fetches pages for actual ChatGPT answers and reaches 55% of the web. Blocking the trainer is a choice. Blocking the search bot is self-sabotage.
The check takes minutes: read your robots.txt, then your Cloudflare bot settings, then ask whoever owns the CDN whether anything filters bots at the edge. If you want the strategic split, it looks like this:
User-agent: OAI-SearchBot
Allow: /
User-agent: PerplexityBot
Allow: /
User-agent: ClaudeBot
Allow: /
# Optional: block training only
User-agent: GPTBot
Disallow: /
This is the tweak with the most dramatic documented payoff:
That site went from 18 to 163 citations in about two weeks, and the biggest lever was discovering their CDN silently blocked several AI crawlers at the edge. As one commenter put it, it's the only AEO fix that can take a brand from invisible to visible without writing a single new sentence.
2. Baseline Before You Touch Anything
Run 10 to 20 real buyer questions through ChatGPT, Perplexity, and Gemini in a logged-out session and write down who gets cited. Half a day, and it's the difference between "we improved" and "we think we improved." Our prompt discovery guide covers building the panel properly, but even a scrappy version beats none. Log four things per prompt: were you mentioned, were you cited with a link, what position, and which URL of yours got used. Run each prompt a few times and score the typical outcome, because single runs lie. Total cost: one spreadsheet and half a Friday evening.
3. Ship llms.txt
Fifteen minutes: a plain-text file telling AI systems what your site is and where the important pages live, per the llms.txt spec. Around 850 sites have adopted it, including Anthropic, Vercel, and Cloudflare. Real talk: our own 90-day, 12-site test found it's no rocket fuel. It's cheap insurance that costs less time than reading arguments about it.
4. Pull Your Pricing Out of JavaScript
If your pricing renders client-side, most AI crawlers see a blank div, and the network-traffic analysis of ChatGPT's fetching showed what happens next: the engine quotes a review site's version of your pricing instead. Server-render the numbers buyers ask about most. Same goes for spec tables, feature grids, comparison widgets, and integration lists: anything a buyer would ask an engine about should survive a curl with JavaScript off. The test is exactly that crude, by the way. Curl the page, search the output for your own price. If it's not there, the engines aren't seeing it either. Friday night, one template fix, permanent effect.
200+ SaaS teams already track their AI citations.
They know exactly when ChatGPT mentions their brand, and when it stops. Do you?
See plans→
Saturday: The Content Pass (the real work)
5. Put the Answer in the First 150 Words
The single most-repeated rule from practitioners who track citations for a living:
"content that gets cited in ai overviews follows 3 simple rules: 1. direct answer in the first 150 words 2. every section tied to a real search question 3. every section complete on its own here is the content brief frame..." — @vibemarketersHQ, June 2026
Direct answer up top, every section tied to a real question, every section complete on its own. Delete the "In this comprehensive guide, we'll explore..." opener from every page you touch. Nobody misses it, least of all the model deciding whether your page contains an answer.
6. Turn Your H2s Into Questions (With One-Line Answers)
Rewrite section headings as the questions buyers actually ask, then answer in the first sentence underneath, then elaborate. Engines match prompts to passages, and a heading that IS the prompt with an answer directly below it is the easiest match in the game.
Before: "Integration Capabilities." After: "Does it integrate with HubSpot?" followed by "Yes, natively, on every plan." That's the whole edit. It matters more than it looks because long queries are where AI answers live: SE Ranking's data shows ten-word queries trigger AI Overviews five times more often than single-word ones. Conversational headings meet conversational prompts.
7. Kill Orphan Pronouns
Practitioners call them orphan pronouns: sentences that say "it handles this automatically" where "it" lives two paragraphs up. Engines lift passages, not pages. A passage that can't survive alone doesn't get lifted. Say the product's name. Say the feature's name. Yes, it reads slightly repetitive to you; you're not the retrieval system.
Before: "It syncs automatically once you connect it." After: "Acme syncs your CRM contacts automatically once you connect HubSpot." The second sentence survives being ripped out of the page and dropped into an answer. The first one dies in transit.
8. One Question Per Page, 800 to 1,200 Words
Here's the counterintuitive one. Ahrefs' AI Overviews research found word count correlates with AI citations at a Spearman of 0.04, which is to say: basically not at all. Grounding value plateaus around 540 words. And a 12-week practitioner study that logged 4,000 citations across 220 pages found 67% of citations went to just 27 pages, which shared a profile: 800 to 1,200 words, one specific question answered completely, and something original on the page. One commenter's "middle child" observation deserves its own plaque: their pages ranking 6th through 12th got cited more than their #1 rankings, likely because the top pages were too dense to excerpt cleanly.
9. Add One Sourced Stat Per Section
The Princeton GEO experiments put numbers on this: adding statistics lifted generative-engine visibility by roughly 40%. A concrete number with a named source turns a paragraph from opinion into evidence, and evidence is what answers are assembled from.
Your best source is your own product data, because a number that exists nowhere else on the internet makes your page the only citable origin for it. That was one of the three traits shared by the top-cited 12% of pages in the 4,000-citation study: something original, a benchmark or data point you couldn't find word-for-word on five other sites.
10. Add an Expert Quote or Two
Same paper, bigger lift: quotation addition came in at +41%, the strongest single method tested. Quotes from real, named humans. Two per priority page is plenty, and between us, this is the tweak nobody does because it requires asking an actual person for a sentence. That's exactly why it still works.
11. Cite Your Sources Inline
Third GEO-paper method, +30% overall, and the fun part: cite-sources more than doubled visibility for pages ranked around fifth position while the top result lost share. Generative engines redistribute attention toward better-evidenced underdogs. Our external links are all nofollow and it changes nothing about this effect; the signal is the citation, not the link equity.
12. Add a TL;DR Block Up Top
A key-takeaways box near the top mirrors the shape of the answer the engine wants to produce. Full disclosure: no controlled study verifies this one; it's practitioner consensus plus the logic of Content-Answer Fit. We wrote up the whole tactic here, and it costs minutes per page. Keep it to three to five bullets, each a complete standalone claim with a number where you have one, formatted the way the engine's own answer would be. Skim an actual AI answer for your target prompt first and mirror its shape.
13. Put a Comparison Table Near the Top
Same honesty label: widely observed, not study-verified. Tables are the densest extractable format when a prompt demands side-by-side answers, and every citation-tracking practitioner I've read swears by them for versus-style queries. Cheap to test on your three most competitive pages. Columns are the options, rows are the criteria buyers actually weigh (price, setup time, team size, support), and the last row is "best for", one honest phrase per option. That last row is the one engines quote.
How often does ChatGPT mention your brand?
Most founders have no idea. The answer might surprise you.
Find out now→
Sunday: Structure and Identity
14. Move Your FAQs Into the Article Body
The standard advice says build FAQ pages. The evidence says otherwise:
That operator's 60-day test moved Q&A content from dedicated FAQ pages into article bodies and watched citation rates jump from 8% to 19%. Standalone FAQ pages are context-free crawler food; the same questions inside a real article give the model a complete frame to quote from. And worth knowing before you spend Sunday on markup: Google deprecated FAQ rich results for general sites, so FAQPage schema no longer buys you SERP real estate. Keep the markup if you have it (LLMs can still read it), but the win is the inline content, not the JSON.
15. Show an Updated Date and Mean It
Freshness wins selection among trusted candidates, and decay is measurable. Two moves: expose a visible updated date backed by dateModified in your Article schema, and refresh on a 21-day rhythm rather than cosmetic weekly touches. The operator from the FAQ test found weekly updates actually made citations less stable; the 21-day cycle steadied them.
16. Give Every Post a Real Author
A named human with a bio page, ProfilePage markup, and sameAs links to their real profiles. Identity signals are among the strongest correlates of AI citation selection, and the full build is in our EEAT and author pages guide, including the plugin noindex trap that quietly voids the whole exercise. The sameAs array wants LinkedIn and X at minimum, plus GitHub or conference pages where they exist. Sunday afternoon, two or three authors, done.
17. Anchor Your Key Terms With DefinedTerm
Pick the 10 to 15 terms your category runs on, give each a glossary anchor with a clean one-sentence definition, and mark them up with DefinedTerm schema so engines resolve your definitions as canonical. The full strategy is our glossary playbook; the weekend version is just your top handful of terms. Each entry opens with the one-sentence "X is a Y that does Z" pattern, self-contained and quotable, because that first sentence is the product and everything after it is packaging.
The Stuff That Didn't Work (Read Before You Overdo It)
Actually, rewind, because a checklist without a failure list is a sales pitch. From the same practitioners who produced the wins above: dedicated FAQ pages underperformed, weekly content churn destabilized citations, maintaining separate "AI-optimized" versions of pages did nothing but double the workload, and spraying schema on every page regardless of type moved nothing. The over-optimization mood is thick enough that it's become a genre of joke:
That's satire, and it lands because half the industry is doing it unironically. There's also a correlation trap worth naming: cited pages tend to have expert quotes and clean structure partly because they were already good, authoritative content. One more from the failure files: chasing citation volume across every AI engine at once. The team that tried it found all-model coverage was a vanity number; focusing on the two engines their buyers actually used drove the traffic. Pick your two before you optimize for five.
These 17 tweaks remove friction and make good content extractable. They don't transmute thin content into citations, and anyone selling that is selling.
Monday Morning: Know If It Worked
So which switches actually mattered? Re-run your baseline prompts two weeks out, same questions, several runs each, and compare against your Friday-night log. Crawler fixes show up in days; content changes take two to three weeks. If you'd rather the whole loop ran itself, that's literally what we built: the Forge Agent applies these patterns across your entire content library, and citation tracking watches every engine so the before/after writes itself.
15 hours a month manually. Or 15 minutes with RankControl.
Track citations, monitor competitors, and fix content gaps across every AI search engine. Automatically.
See how it works→
Seventeen switches, one weekend, evidence attached. Flip the Friday ones tonight even if you skip everything else, because somewhere in your stack there's probably a rule quietly telling the answer engines you don't exist.
