The AEO Audit – What Actually Gets You Cited by AI

An AEO audit measures whether AI answer engines like ChatGPT, Perplexity, Google’s AI Overviews, and Gemini can find your brand, cite it, and recommend it when a buyer describes their problem. A real audit goes past the technical checklist. It scores five layers, and the layer that decides most of your citations is the one nearly every audit barely touches.

I’ll show you the framework, back every claim with data, and run it live against a national brand you know so you can see exactly what it turns up.

What most AEO audits get wrong

Search “AEO audit,” and you’ll find the same thing over and over: a 40-signal technical checklist. AI crawler access, schema markup, llms.txt, heading structure. Tidy, scannable, and mostly beside the point.

Here’s the problem. The checklist assumes AI visibility works like traditional SEO, where you fix the on-page signals and the citations follow. The data says otherwise. Ahrefs analyzed 75,000 brands and found that brand mentions across the web correlate with AI citation rates about three times more strongly than backlinks do, and the mention counts whether or not it carries a link. A separate synthesis by 5W found brand search volume to be the single strongest known predictor of whether AI cites you, outweighing backlinks. Neither of those levers lives on your website, and no checklist item touches them.

Ranking well doesn’t rescue you either. In one study of 40,000 keywords, 88% of the sources cited in Google’s AI Mode did not come from the organic top 10. Another 28% of the pages ChatGPT cites most often have zero Google organic visibility at all. You can own page one and still be absent from the answer.

Then there’s schema, the checklist’s favorite. It earns its place for entity clarity, but it is not the citation driver it’s sold as, and the highest-profile example of “schema as AEO” is now gone. Google removed FAQ rich results in May 2026. The FAQ content still helps your reader. The markup no longer buys you anything in search, and it never bought you AI citations.

So a good audit doesn’t start by counting schema types. It starts by asking a harder question. When a buyer asks AI about your category, are you in the answer, and if not, what would actually put you there?

The audit works in five layers. Here they are, in the order you run them.

Layer 1: Can AI engines actually read your site?

This is pass/fail. It won’t earn you a single citation, but it can erase you from every answer regardless of how good everything else is. This is table stakes, and more brands fail it than you’d guess.

A Vidern study in July 2026 tested the top 1,000 websites and found 40.9% were unreadable to GPTBot, the crawler behind ChatGPT, and nearly one in five, or 18.4%, were completely dark to every AI crawler tested. These are the most visited sites on the internet, and a large share of them have quietly locked the door.

Here’s the part almost every audit gets wrong: AI crawler access is not one switch. The major AI companies split their crawlers into two jobs, training and search, and the two answer to different rules. Blocking OpenAI’s GPTBot stops it from using your pages to train models, but it has zero effect on ChatGPT Search, which runs on a separate crawler called OAI-SearchBot. OpenAI, Anthropic, Amazon, Google, and Apple all separate training access from search access, which means the old “block all AI bots” advice is now actively harmful to visibility.

You can block the training bots to protect your content and keep every AI-search citation. Or you can block the wrong bot by accident and delete yourself from the fastest-growing referral channel there is. Google-Extended is the classic trap. The name suggests it governs how Google’s AI uses your site, so anxious teams block it to defend their traffic. But it only controls one thing: whether your content trains Google’s Gemini models. It has no effect on Googlebot, your rankings, or whether you show up in AI Overviews. The only outcome is opting out of model training, which is rarely what they meant to do.

Then there’s the silent version of this failure. Vidern found that 17.6% of top sites allow GPTBot in their robots.txt but return a 403 Forbidden when it actually shows up. Their firewall or CDN is blocking traffic their published policy permits, and they have no idea they’re invisible. Your robots.txt says “come in,” and your server says “no.” Nobody on your team is looking at the difference.

Now the access killer that catches the most sophisticated teams: rendering. If your pages build their content in the browser with JavaScript, the way many modern frameworks do by default, the AI engines may never see that content at all.

Here’s the finding that should reframe how you think about this. A joint analysis by Vercel and MERJ tracked more than 500 million GPTBot fetches and found zero evidence of JavaScript execution. The major AI retrieval crawlers, including GPTBot, ClaudeBot, and PerplexityBot, fetch your raw HTML, read what’s already in it, and leave. They sometimes download your JavaScript files (GPTBot in roughly 11.5% of requests, ClaudeBot in about 24%) but never run them. There is no second rendering pass the way Google has.

That produces a genuinely strange situation: the same page can rank on page one of Google and be a blank shell to ChatGPT, Claude, and Perplexity at the same time. Googlebot renders JavaScript in a second wave, so client-side content ranks in classic search. The AI crawlers don’t, so that same content is invisible in the answers. The one exception worth naming is Gemini, which reads JavaScript because it runs on Google’s rendering infrastructure, so a client-rendered page can surface in Gemini while staying dark to the others.

The test takes one line. Open the page and view source- the actual raw source rather than the rendered developer-tools panel- and search for a sentence you can see on the page. If your body copy and schema aren’t in the raw HTML, the AI crawlers can’t see them either.

One clarification, because teams get this wrong in both directions. The problem is client-side rendering rather than JavaScript itself. A modern JavaScript framework that pre-renders your pages to static HTML at build time puts the words right there in the raw response, where the crawlers can read them. Gatsby’s default mode does exactly this. It renders the whole site to static HTML at build time, and the React that runs afterward only adds interactivity without hiding the content. A “JavaScript site” built this way is fully visible to AI, while a pure single-page app that paints its content in after the page loads is the one that goes dark.

We see this in the wild. One of our clients runs a headless site built on Gatsby. Because Gatsby pre-renders to static HTML, its content sits in the raw HTML the AI crawlers read, even though it’s a JavaScript-framework site. Being headless and React-based didn’t doom it. The build-time rendering is what keeps its content visible. The one catch is that static generation only covers what exists at build time. If a page fetches something in the browser after it loads, that piece goes invisible again, which is exactly why the view-source test is run per page rather than once for the whole site.

What the audit does at this layer:

• Checks each major AI crawler by name, training and search variants separately, in both robots.txt and the live server response, since the two can disagree.
• Confirms the retrieval bots that drive citations are allowed, whatever you decide about the training bots.
• Verifies your body copy and schema live in the raw HTML rather than in JavaScript the AI crawlers won’t execute.
• Flags anything returning a 403 to a bot your own policy permits, and reads server logs to see what the AI crawlers actually fetched and got back.

Passing this layer doesn’t make you visible. It just means you’re eligible. Fail it, and nothing else in this audit matters, so it goes first.

Layer 2: Where do you show up right now, and on which engine?

You can’t fix what you haven’t measured, and here’s the trap: you cannot measure AI visibility with one number. Every “AI visibility score” that collapses the engines into a single figure is telling you something false.

The engines don’t share a brain. Across 680 million citations, only about 11% of domains were cited by both ChatGPT and Perplexity, so the source pools are almost completely separate. They don’t even cite at the same rate. One 2026 study of 34,234 AI responses found a 46-fold gap in how often platforms cite brands at all: ChatGPT mentioned brands 0.59% of the time, Perplexity 13.05%, and Grok higher still. Perplexity also packs in far more sources per answer, about 21.9 citations per response versus ChatGPT’s 10.4. A brand can own Perplexity and be a ghost in ChatGPT. Average those into one score and you’ve hidden the only thing worth knowing.

That’s why the baseline is measured per engine, and why it comes second. You run it after you’ve confirmed the engines can read you, and before you spend a dollar fixing anything. It answers “where do we actually stand,” and it becomes the benchmark every later fix is measured against.

For each engine your buyers use, the audit tracks four things:

• Mention rate: how often you come up at all when someone asks about your category.
• Citation rate: how often that mention carries a link back to you, which is a different and rarer event. AI will talk about your brand without sending anyone to you.
• Sentiment: what the answer says about you. Being present but described badly is its own problem.
• Share of voice: you versus named competitors in the same answers. This is the number that lands in a boardroom.

The inputs matter as much as the metrics. You run real buyer prompts, the messy, problem-shaped questions people actually type, rather than vanity queries with your brand name already in them. Searching your own name proves nothing. The point is to see whether you surface when the buyer hasn’t thought of you yet.

One thing the baseline can’t come from is your analytics. AI referral traffic is badly undercounted. Depending on the study, 35% to 70% of AI-driven visits arrive with no referrer and get filed as Direct. Perplexity often isn’t recognized as an AI source at all and lands in generic Referral. Google’s AI Overviews get counted as ordinary organic search. Your analytics “AI traffic” number is a floor rather than a measurement. The baseline comes from running the prompts yourself and, where you can, reading server logs to see which engines fetched which pages.

And you cover the surfaces your buyers cover. Buyers now check an average of 2.4 platforms before they trust a purchase decision, and trust in AI search dropped from 82% to 54% in a single year, which means people are cross-checking more, not less. One engine is a sample size of one. There’s also Google’s AI Overviews to weigh separately, because that surface only appears on roughly a third of searches. Part of the baseline is simply how often the AI answer shows up at all for your category.

What this layer produces is a per-engine scorecard: for each platform, your mention rate, citation rate, sentiment, and share of voice against a named competitor set, on a fixed prompt list you can re-run. That last part, a fixed and repeatable prompt list, is what turns a one-time snapshot into a benchmark, which matters more than it sounds for reasons that land in Layer 5.

Layer 3: Is your content built to be extracted?

This is the layer the checklists are least wrong about, and it still isn’t the lever. It decides whether an AI can lift a clean, correct answer straight off your page. Get it right and you’re quotable. Get it right and also have the authority from Layer 4, and you get cited. One without the other doesn’t work.

Start with the mental model, because it’s different from SEO. AI systems evaluate content at the passage level rather than the page level. They’re pulling the one paragraph that answers the question, not ranking the whole URL. What you want is a page whose passages each stand on their own and say something true in the fewest words possible, so a machine can lift any one of them cleanly.

That’s why answer-first structure works. Open each section with the direct answer and follow it with the support. A machine skimming for a liftable sentence finds it immediately, and so does a human. The same move serves both readers, which is the only kind of optimization worth doing.

Two things separate content that gets extracted from content that just sits there.

Original evidence. AI engines cite the original data source rather than the page that quotes it, so if your whole article is built on other people’s statistics, the citation flows to them and not to you. This is the single most actionable finding at this layer. The brands getting cited are the ones publishing something only they can: their own numbers, their own test, their own survey, their own named framework. Princeton’s research on generative engine optimization found that adding authoritative signals like citations, statistics, and direct quotes lifts AI visibility by 30 to 40%. Aim to be the source that others summarize.

Every page type counts, not only the blog. Product pages, pricing pages, documentation, and customer stories all get cited by AI engines, so a complete audit covers every page type that answers a question your buyer asks. Most audits scan the blog and stop. When a shopper asks an AI what a product actually includes, the engine wants a clean, factual product or FAQ page to lift from, and if yours buries that answer in marketing copy, it reaches for a third-party review site instead. That’s a citation you handed away.

Schema lives here too, in its proper size. It helps a machine parse your entities and disambiguate who you are. It does not manufacture citations, and after Google dropped FAQ rich results in May 2026, one of its headline use cases is gone. Mark up what genuinely clarifies your identity, and don’t mistake it for the work.

What the audit does at this layer:

• Checks whether key pages answer their core question in the first sentence rather than the fifth paragraph.
• Maps your pages against the real questions buyers ask at each stage, and flags the questions no page answers.
• Identifies where you’re quoting others’ data instead of publishing your own, the highest-value gap here.
• Confirms every page type that answers a buyer question is audited, not only the blog.
• Verifies entity schema is accurate and current, and retires anything that was only there for deprecated rich results.

Done well, this layer makes you extractable. Whether that extraction turns into an actual citation depends almost entirely on the next layer.

Layer 4: Are you cited where it actually counts?

This is the layer. If Layers 1 through 3 make you eligible and extractable, this is where citations are actually won, and it’s the layer the checklists can’t help you with, because none of the work happens on your website.

The data here isn’t subtle. Brand mentions across the web correlate with AI visibility at 0.664, roughly three times stronger than backlinks at 0.218. Brand search volume is the strongest single predictor anyone has isolated, ahead of backlinks. Distributing your content and mentions across a range of publications can lift AI citations by up to 325% compared with publishing only on your own site. The pattern is consistent across every study: AI cites the brands that show up, by name, in a lot of trusted places, not the brands with the cleanest schema.

Now the uncomfortable part. A lot of those trusted places are ones you cannot buy your way into. 67% of the pages ChatGPT cites most often are effectively off-limits to brand marketing, including Wikipedia, government and educational sites, major news outlets, and reference sources. You earn your way adjacent to them. This is why the honest framing of AEO isn’t “optimize your page.” It’s closer to public relations than to SEO, the work of earning your name into the sources the engines already trust. Some people in this space have started calling it machine relations, and the label fits. Because each engine pulls from a different pool of sources, the only way to build citation share across all of them is through earned coverage in the publications they each rely on.

Every category has a short list of those sources, and the concentration is severe. In B2B SaaS CRM, a single publication, TechRadar, accounts for 8.86% of all category citations. In healthcare, Google’s AI Overviews lean on a tight set: NIH, Healthline, Mayo Clinic, Cleveland Clinic, and ScienceDirect. Win placement in the handful of sources that own your category and you’ve moved the needle more than a year of on-page tweaks would.

Two more signals belong here.

Entity consistency. The engines need to know, without ambiguity, who you are: one entity, one set of facts. When your company name, founding details, location, and leadership read differently across your site, LinkedIn, Crunchbase, and Wikidata, you’ve introduced doubt into the exact system that decides whether to cite you. The audit checks that your entity graph tells one clean story everywhere the engines look.

Video. This one’s underused and worth acting on fast. Ahrefs’ study of 75,000 brands found that brand mentions in YouTube video titles and transcripts were the single strongest correlating factor with visibility in Google’s AI Overviews, stronger than any on-page signal studied. If your category gets researched on YouTube and your brand name isn’t in the titles and transcripts, that’s a citation channel sitting untouched.

What the audit does at this layer:

• Maps the specific sources AI engines cite for your category, per engine, and measures whether you appear in them and how you’re characterized.
• Benchmarks your branded search volume and web mention frequency against named competitors, the two signals that predict citations best.
• Audits entity consistency across your site, LinkedIn, Crunchbase, Wikidata, and other reference sources the engines cross-check.
• Identifies the earned-media and roundup placements that would move you into the sources you’re currently missing.
• Checks your presence in the video and community surfaces your category actually gets researched on.

This is slow work. It doesn’t produce a same-week screenshot, and that’s exactly why most audits skip it in favor of schema they can check off in an afternoon. But it’s the work that decides whether AI recommends you, so a real audit puts the majority of its findings, and your budget, right here.

Layer 5: How do you keep it from decaying?

Here’s what nobody selling you a one-time AEO audit wants to say out loud: the result expires quickly. A citation audit is a photograph of a moving target, and the target moves monthly.

Roughly half of the pages AI engines cite change every single month, and nearly 6 in 10 cited pages appear once and never return. Sit with that. The source an engine quotes for your category today has a coin-flip’s chance of being a different source next month. A visibility win you earned in March can quietly evaporate by May with no change on your end, because the engine simply re-drew from a different pool.

The churn isn’t random noise you can ignore, either. It’s large and abrupt. ChatGPT’s citation share from Reddit collapsed from around 60% to about 10% in the span of a single mid-September 2025 window. Overnight, every brand whose AI visibility rested on Reddit threads lost most of it. Any strategy built on one source or one platform is fragile by design, since the ground underneath it can shift regardless of how sound the strategy is.

This is the argument against treating an AEO audit as a certificate you frame and forget. A one-time audit tells you where you stood on one day, in a system that won’t hold that shape. That makes it a useful starting point but a poor stopping point.

So the fifth layer is a decision about cadence rather than a checklist item. The audit sets up three things that only pay off over time:

• A fixed prompt panel you re-run. The same buyer prompts from Layer 2, run on a schedule, so your mention rate, citation rate, and share of voice become a trend line instead of a one-off number. The trend is the signal, and a single reading tells you very little.
• Freshness discipline on the pages that matter. The pages you most want cited need a real update rhythm, which means genuinely revisiting the answers, data, and claims the engines are lifting rather than doing busywork. Stale sources get dropped.
• A watch on the category’s citation landscape. The sources that own your category this quarter aren’t guaranteed to own it next quarter. Part of the ongoing work is noticing when a new publication or forum starts feeding the engines in your space, and earning your way in before your competitors do.

That reframes what you’re actually buying. It’s a baseline plus an instrument rather than a one-time audit. The first audit establishes where you stand and what to fix. The recurring measurement proves whether the fixes moved anything, and catches decay while it’s still cheap to reverse. In a system this volatile, the measurement is the product.

This is the honest reason we run AEO as an ongoing engagement rather than a one-time report. It isn’t about keeping a retainer alive. The alternative is optimizing once, walking away, and having no idea three months later whether you’re still in the answer or quietly gone.

What an AEO audit looks like in practice: Warby Parker

Theory is easy. Here’s the framework run against a brand you know. I gave ChatGPT, Perplexity, and Google’s AI Overview the same buyer question, “Warby Parker vs. Zenni,” and watched what each one did. Warby Parker is a strong test case: 673,000 branded searches a month and a reputation built on marketing. If any brand should own its own story in AI answers, it’s this one. Here is what the audit turned up.

Nobody cited Warby Parker’s own website. This is the finding that matters most, and it showed up on every engine. Google’s AI Overview built its entire comparison from Reddit threads, YouTube reviews, Healthline, Medical News Today, and a handful of affiliate blogs. Perplexity leaned on a single third-party source, Healthline. ChatGPT wrote a full comparison table and cited nothing at all, pulling from memory and sending no one anywhere. Across three engines answering a question about Warby Parker, not one of them quoted warbyparker.com. The brand’s story was told entirely by other people. That is Layer 4 in a single screenshot: when you’re absent from the sources the engines trust, they build your narrative without you.

One engine told shoppers about a program that no longer exists. Warby Parker announced in August 2025 that it would end its signature Home Try-On program, and it did so by the end of that year. ChatGPT had this right and noted the program had ended. Google’s AI Overview told shoppers Warby Parker offers a popular free Home Try-On program shipped to their home. That is wrong, and it’s wrong for a specific reason: the AI Overview pulled the claim from older third-party pages instead of Warby Parker’s current site. A buyer reading that answer could show up expecting a service that’s gone. This is what Layers 3 and 4 exist to catch. The current fact has to live on Warby’s own extractable pages and work its way into the third-party sources the engines lean on, or the engines keep repeating the old one.

The engines agreed on the story, and the story hands a whole segment to the competitor. All three framed Warby Parker the same way: physical stores, professional fitting, a curated selection, and prices starting around $95. They framed Zenni as the value pick at $6.95 with a huge catalog. That consistency is good news for the brand. The problem is the verdict. ChatGPT put it plainly, “Zenni wins on value, and it isn’t close,” and the AI Overview drew the same split. For any buyer who leads with price, the engines route to the competitor by default. Warby Parker can’t win that shopper on price, but right now it isn’t contesting the framing at all.

The three engines behaved nothing alike. ChatGPT cited zero sources, Perplexity cited one, and the AI Overview cited a dozen across Reddit, YouTube, and review blogs. A single blended “AI visibility score” would have averaged all of that into a meaningless number. Read per engine, it tells Warby Parker exactly where the work is: earn a place in the review sites and Reddit threads that feed Google’s AI Overview, and get the current facts into the sources Perplexity and ChatGPT draw on.

One honest limit on this cut. Naming both brands in the prompt tests how the engines frame a comparison, and not whether Warby Parker surfaces when nobody names it. The sharper follow-up is the brand-free prompt, something like “best places to buy glasses online” or “affordable prescription glasses for a strong prescription,” which shows whether Warby Parker is in the answer at all before the buyer thinks of it, and who else is in the room. That’s the next prompt set to run, and it’s exactly what a full audit covers.

AEO audit FAQ

What is an AEO audit?

An AEO audit measures whether AI answer engines can find, cite, and recommend your brand when a buyer asks about your category. It checks five layers: whether the engines can read your site, where you currently show up per engine, whether your content is built to be extracted, whether you’re cited in the sources the engines trust, and how you’ll keep that position from decaying. The output is a per-engine baseline and a prioritized list of what to fix.

What’s the difference between an AEO audit and an SEO audit?

An SEO audit optimizes for rankings and clicks, while an AEO audit optimizes for citations. They overlap on technical hygiene, since both need a crawlable, readable site, but they diverge on what wins. In one 40,000-keyword study, 88% of the sources cited in Google’s AI Mode didn’t come from the organic top 10, so a clean SEO audit can pass while your AI visibility fails. The AEO audit spends most of its attention on off-site authority, which a traditional SEO audit barely touches.

Is AEO the same as GEO?

Mostly, yes. They’re two names for the same problem: getting surfaced inside AI-generated answers instead of blue links. In practice, AEO (answer engine optimization) emphasizes being the cited answer in engines like ChatGPT, Perplexity, and Gemini, while GEO (generative engine optimization) is used as the broader umbrella term. Don’t get hung up on the label. The work is the same.

Does an AEO audit replace your SEO?

No. It layers on top. The technical foundation an SEO audit checks, including crawlability, site health, and clean structure, is table stakes for AI visibility too, since the engines can’t cite a page they can’t read. AEO adds the authority, entity, and per-engine measurement work that traditional SEO never covered. Rather than replacing SEO, you’re extending it into the surfaces where buyers now research.

How often should you run one?

Quarterly at a minimum, and monthly if your category is competitive. The reason is volatility. Roughly half of AI-cited pages change every month, and nearly 6 in 10 appear once and never return. A one-time audit tells you where you stood on a single day in a system that won’t hold that shape. The first audit sets the baseline, and the recurring ones prove your fixes worked and catch decay while it’s cheap to reverse.

How long does an AEO audit take, and what does it cost?

Scope drives both. A thorough manual audit of a mid-sized site runs roughly 40 to 60 hours over about two weeks when done in-house. The technical and content layers move fastest, while the authority and citation-landscape mapping is where the real time goes, because it means analyzing the specific sources each engine trusts in your category. Agency pricing varies with the number of engines, competitors, and page types in scope, and a focused baseline costs a fraction of a full multi-engine audit with an ongoing measurement program.

The audit we ran on ourselves

I’ll be straight with you: we ran this exact five-layer audit on our own agency, and we didn’t love what it turned up. For all the right technical signals, we weren’t showing up when people asked AI to recommend an agency in our market, because we were absent from the third-party sources the engines actually pull from. It was the same gap I just described in Layer 4. Knowing the framework doesn’t exempt you from it.

We’re fixing it in public, and we’ll publish that journey as we go, because “here’s what worked on us” is worth more than any checklist.

That’s the real point of an AEO audit. It gives you a clear read on which of your instruments are actually playing in the answer, along with an honest plan to get the rest of them heard. A score to frame was never the goal.

If you want to know where your brand stands across ChatGPT, Perplexity, and Google’s AI Overviews, and what would actually move you into the answer, that’s the AEO audit we run. Get in touch, and we’ll show you where you stand.

Sources

Every statistic in this article traces to the studies below. Where a primary study is reported by a secondary source, both are noted.

1. Ahrefs analysis of 75,000 brands on brand mentions versus backlinks (0.664 vs. 0.218) and YouTube title and transcript correlation. Reported by Omnibound, “Generative Engine Optimization Statistics (2026).” https://www.omnibound.ai/blog/generative-engine-optimization-statistics
2. 5W, “The State of AI Citations 2026” (brand search volume as strongest predictor; TechRadar at 8.86% of CRM citations; healthcare source set; the ChatGPT and Reddit shift). https://www.5wpr.com/research/state-of-ai-citations-2026/
3. Moz 40,000-keyword AI Mode study and the 11% ChatGPT and Perplexity domain overlap. Reported by AuthorityTech. https://authoritytech.io/curated/ai-citation-11-percent-platform-overlap-per-engine-audit-2026
4. ChatGPT most-cited pages: 67% off-limits, 28% with zero Google organic visibility, and per-response citation counts. QuickSEO. https://quickseo.ai/blog/chatgpt-vs-perplexity-for-ai-visibility-in-2026-citations-traffic-and-conversion-compared
5. Vidern top-1,000 crawler-access study, July 2026 (40.9% unreadable to GPTBot; 18.4% dark to all crawlers; 17.6% allow in robots.txt but return 403). Reported by AuthorityTech. https://authoritytech.io/curated/publisher-google-block-crawl-access-ai-brand-strategy-2026
6. AI crawler training-versus-search split and access decision matrix. Digital Applied. https://www.digitalapplied.com/blog/ai-crawler-access-control-2026-robots-llms-txt-decision-matrix
7. Google-Extended scope (training only, with no effect on Googlebot or AI Overviews). Parse.gl. https://parse.gl/blog/robots-txt-ai-crawlers-configuration
8. Vercel and MERJ, “The Rise of the AI Crawler,” 500M+ GPTBot fetches with zero JavaScript execution. https://vercel.com/blog/the-rise-of-the-ai-crawler
9. JavaScript rendering gap detail (GPTBot roughly 11.5%, ClaudeBot roughly 23.84% download but do not execute; Gemini as the exception). Averi. https://www.averi.ai/blog/javascript-rendering-gap-ai-crawlers
10. View-source and raw-HTML rendering test. SourceCited. https://sourcecited.com/rendering-crawlability.html
11. The 46-fold brand-citation gap across platforms and the YouTube correlation in AI Overviews. Leapd. https://www.leapd.ai/blog/ai-visibility/how-chatgpt-google-ai-overviews-and-perplexity-source-information-in-2026
12. Perplexity Reddit citation share and AI Overviews appearing on roughly a third of searches. Otterly.ai, “The AI Citations Report 2026.” https://otterly.ai/blog/the-ai-citations-report-2026/
13. Princeton GEO study (a 30 to 40% visibility lift from authoritative signals). Reported by PingPrime. https://www.pingprime.ai/en-be/blog/chatgpt-search-ai-overviews-perplexity
14. Buyer trust in AI search falling from 82% to 54% and buyers checking 2.4 platforms before purchase. AuthorityTech. https://authoritytech.io/curated/ai-search-trust-dropped-measurement-stack-brand-mentions-2026
15. Citation churn (about half of cited pages change monthly; roughly 6 in 10 appear once) and page-level extraction. AirOps, “AEO Audit Checklist.” https://www.airops.com/blog/aeo-audit-checklist
16. Manual AEO audit effort (40 to 60 hours over about two weeks). RevvGrowth. https://www.revvgrowth.com/aeo/aeo-audit-checklist
17. GA4 AI-traffic undercounting (35% to 70% filed as Direct; Perplexity not in the native AI channel; AI Overviews counted as Organic). AuthorityTech and Vynce Digital. https://authoritytech.io/blog/ai-traffic-attribution-how-to-track-chatgpt-perplexity-gemini
18. Gatsby rendering options: SSG is the default, and the site is pre-rendered to static HTML at build time, with React hydration adding interactivity only. Gatsby documentation. https://www.gatsbyjs.com/docs/conceptual/rendering-options/
19. SSG and SSR versus client-side rendering for AI crawlers: React through Gatsby produces server-rendered HTML crawlers can read, while a pure SPA goes dark. GReadme, “Client-Side Rendering and SEO.” https://www.greadme.com/blog/seo/seo-issues-with-client-side-rendering-complete-guide