Team Madcraft
Olivia has more than five years’ experience leading SEO strategy for brands across Ireland and the US, with expertise in technical SEO, content and search intent, and a recent focus on Answer Engine Optimisation, AI Overviews and LLM visibility.
You rank on page one for a term that matters to your business. Great! Now, ask ChatGPT or Perplexity the same question: a competitor gets named while you are not mentioned at all.
That is the moment a business starts asking how to get cited in AI search. The assumption is that the content needs rewriting. In most of the audits we run, the content is fine. The AI system never received it.
We have spent the last eighteen months building, auditing and measuring Irish sites for answer engine visibility. Most AI crawlers read the raw HTML your server returns and stop there. If your text only appears once JavaScript has run, they see an empty page. That is a development ticket, not a content brief.
•
•
•
•
•
•
We audit what AI systems can actually read on your site, then fix the access, structure and content in the order that pays back.
Three things have to be true, in order: an AI system has to be able to read your page, it has to be able to retrieve a useful passage from it, and it has to have a reason to prefer your passage over everyone else’s.
Almost every AEO checklist online starts at step three. That is why so much AEO work produces nothing. Rewriting a page for clarity does not help if the crawler never received the text, and adding structured data does not help if the passage underneath it says the same thing as forty other pages.
We work through those three steps in that order on every engagement. The order is the method.
Because the content is not in the HTML the server sends back. Most AI crawlers request a page, read the raw HTML response, and stop. They do not execute JavaScript.
Joint research by Vercel and MERJ analysed more than 500 million GPTBot fetches and found no evidence of JavaScript execution at all. GPTBot downloaded JavaScript files in around 11.5% of requests and ClaudeBot in around 23.8%. Neither ever ran them.
That finding has held. Independent crawler monitoring re-confirmed it in June 2026 across GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, PerplexityBot, Meta-ExternalAgent and Bytespider. There is a resource logic behind it: these crawlers work to tight timeouts, and rendering JavaScript at their volume would be expensive. Skipping it keeps retrieval fast and cheap, so it is unlikely to change soon.
Gemini is the only meaningful exception, and that is what makes this so easy to miss. Googlebot renders JavaScript in a separate, later stage, and Gemini inherits the same Web Rendering Service. So a site built with client-side rendering can rank perfectly well in Google Search, look healthy in Search Console, and still return a blank page to ChatGPT, Claude and Perplexity.
This affects more Irish business websites than people expect. It is not only single-page applications. Common patterns that cause it include:
The last one matters more than it sounds. If your structured data is added by JavaScript, the systems most likely to benefit from it are the ones least likely to see it.
One distinction to call out
Content that is missing from the HTML is a different problem from content that is present but hidden. AI crawlers read the raw response and do not apply your stylesheets, so text held at zero opacity by a scroll animation until a script runs is still perfectly readable to them. That is a rendered-snapshot issue for Google, not an access issue for ChatGPT.
It also works the other way. Content does not have to be visible prose to count. Inline JSON, structured data and server-rendered component payloads all sit in the initial response and are readable. What these crawlers miss is specifically what the browser builds after the page loads.
The two faults need different fixes, so it is worth knowing which one you have before anyone opens a development ticket. The test in the next section tells you.
Use view-source or a command line request, not your browser inspector. The browser inspector shows the rendered DOM after JavaScript has run, which is exactly the view an AI crawler does not get.
Two checks settle almost every disagreement about this:
If the text is missing, the fix is a development task, not a content task. Server-side rendering, static generation or prerendering of the main content blocks puts the text in the initial HTML for every crawler at once. Our web development for AI search service builds this in at architecture stage, because retrofitting it after launch is always slower and more expensive.
They retrieve at passage level, not page level. Google’s own guidance describes AI Overviews and AI Mode as running on the same crawl, index and quality systems as ordinary Search, using retrieval and query fan-out to assemble an answer from multiple sources.
Query fan-out means one question is broken into several related searches. Your page can contribute to an answer without matching the original wording, as long as one section of it cleanly answers one of the sub-questions.
The practical consequence is that each section has to stand on its own. A passage that only makes sense after reading the two sections above it is a weak candidate. The patterns that consistently retrieve well:
The most common mistake we see on otherwise strong pages is burying the answer. If the response to the H2 question sits in the fourth paragraph, the retrieval layer often never reaches it.
By ranking and being useful. Google published its AI optimization guide in May 2026 and was direct about it: there are no additional requirements to appear in AI Overviews or AI Mode, and no special AI markup.
That guidance is more useful than it first appears. It means the work is the same work: crawlability, indexation, snippet eligibility, relevance and content that adds something. Traditional ranking remains the strongest single predictor of being cited in an AI Overview, which is why answer engine optimisation and SEO should run as one workstream rather than two budgets.
It also means AI Overviews behave differently from ChatGPT and Perplexity in one important way. Gemini can render JavaScript. The others cannot. A site with a rendering problem may appear in AI Overviews and be entirely absent from every other answer engine, which is a genuinely confusing signal to read if you are only checking one surface.
Content that contains something a model cannot get anywhere else. Google’s guidance calls it non-commodity content. In practice it means the page carries information that could not have been assembled from other pages.
What earns citations in our tracked work:
What gets skipped: restated definitions, roundups of other people’s advice, and pages built to cover every possible phrasing of the same query. Google’s guidance warns specifically about producing large volumes of near-identical pages, and treats it as scaled content abuse when it is done to manipulate visibility.
Several of the most heavily marketed AEO tactics produced nothing measurable, and independent research now backs that up.
None of this makes technical work optional. It moves the effort to the part that is actually load-bearing: whether the content is readable, useful and distinctive.
Keep using it. Google retired the FAQ rich result on 7 May 2026 and removed the last of the related Search Console support in August 2026. That changed how a listing looks in Google. It did not change what the markup does.
FAQPage markup hands an answer engine a clean question and answer pair, with no ambiguity about where one ends and the next begins. That is the shape retrieval systems work in. We still deploy it on any page with a genuine visible FAQ section, alongside Article and BreadcrumbList.
The only thing that changed is what you should expect back from it. Add it for retrieval, not for a richer Google listing, and never add it to a page with no visible FAQ.
One of the clearest examples we have seen comes from a purpose-built commercial landing page for a national Irish home-security client. This was not an existing page that we lightly reworked for AI search. We designed it from the outset to perform in both traditional search and answer engines.
The structure was deliberate: a Quick Facts block immediately after the introduction for scannability, question-led headings addressing individual decision-stage queries, concise answers that could stand independently when retrieved out of context, and an FAQ section based on the questions people were actually asking. We also attached specific facts and entities directly to the relevant answers rather than leaving them buried elsewhere on the page.
The page first appeared in our AI citation tracking in February 2026. It has since been cited across the exact types of questions it was built to answer:
Citation share on the tracked prompts:
| Tracked prompt | Citation share |
|---|---|
| Which alarm company has the most customers in Ireland? | 14.99% |
| Which alarm company has the fastest response time in Ireland? | 8.11% |
| Best home alarm systems Ireland | 6.65% |
What this reinforced for us is that AEO success is rarely the result of one optimisation. Everything from the technical foundations laid during our initial audit and site clean-up to the structure, layout and specificity of the page played a role. The technical work made the content accessible; the answer-led structure made it retrievable; and the specific facts and entities gave AI systems something worth citing.
AI visibility can be tracked, but not with the precision people expect from rankings. Results vary by model, prompt wording, location and date, and the same question asked twice can return different sources.
What we treat as reliable enough to act on: citation presence for a defined set of priority questions, checked on a consistent schedule; growth in branded search; referral sessions from AI platforms in GA4; and assisted conversions. Google reports clicks from AI Overviews and AI Mode inside the Web search type in Search Console rather than as a separate report, so there is no clean split available there.
What we do not treat as reliable: any single vendor visibility score presented as an absolute. It is useful for direction of travel and useless as a target.
If you want to know where you stand before committing budget, an SEO audit that includes a rendering check will answer the first and most expensive question in about a day. Our earlier guide on how SEO and AEO fit together covers the strategic split in more detail.
Getting cited in AI search takes three things in order: a page an AI crawler can read in the raw HTML, a passage that answers one question on its own, and content specific enough that a system prefers it to a generic summary. Madcraft works through those three in that order rather than starting with just content.
The most common cause is that your main content only appears after JavaScript runs, so the crawler receives an almost empty page. Madcraft checks this by reading the raw HTML response rather than the browser view, which usually settles the question in minutes.
No. Google has confirmed it is not required, and independent studies have found almost no AI crawler traffic to the file. Madcraft treats it as optional and prioritises rendering and content quality first.
Yes, wherever the page has a genuine visible FAQ section. Google retired the FAQ rich result in May 2026, so it will not change how your listing looks, but the markup still gives answer engines a clean question and answer pair to retrieve. Madcraft deploys it alongside Article and BreadcrumbList.
No. Google’s AI features run on Googlebot and its rendering service, while ChatGPT, Claude and Perplexity use their own crawlers that read raw HTML only. Madcraft checks visibility separately for each, because a site can perform in one and be absent from the other.
It depends on how often the page is recrawled and how competitive the question is. Madcraft tracks a defined set of priority questions from the point of the fix so the change is visible rather than assumed.
Smaller businesses are cited regularly, particularly for specific service and location questions where general publishers have nothing useful to say. Madcraft focuses client content on the questions where a specialist answer beats a generic one.
