The most important lesson from Wix Studio and Peec's large AI citation study is not that one content format has “won” AI search. It is that AI systems pull from different kinds of content depending on what the user is trying to accomplish.
That changes GEO strategy. Instead of asking, “What content does ChatGPT prefer?” marketers should map the buyer's task first, then build the owned and third-party evidence most likely to support that task across AI systems.
GEO should be organized around intent and evidence, not a single “AI-friendly” content format.
The study's strongest pattern was that prompt intent predicted citation format more strongly than either industry or AI model.
What Wix and Peec actually studied
Wix Studio's AI Search Lab and Peec analyzed 75,000 AI answers containing 1,056,727 citations across ChatGPT, Google AI Mode, and Perplexity. The researchers grouped prompts into four intents—informational, commercial, navigational/local, and transactional—and examined results across eCommerce, health and wellness, home repair, professional services, and SaaS.
The study therefore did more than count which websites appeared most often. It asked a strategically useful question: when the user's job changes, does the type of content cited by AI change too? The answer was yes, and the differences were substantial.
AI answers analyzed across three major generative search experiences.
Total citations classified by content type, intent, industry, and model.
Prompt-intent groups used to explain how citation behavior changes.
The study's most important finding: intent beats format-first thinking
Across all intents and industries, listicles represented 21.88% of citations, articles 16.68%, and product pages 13.66%. Those overall averages are interesting, but they hide the behavior that matters most to a GEO strategist.
When users asked informational questions, articles jumped to 45.48% of citations. When users asked commercial questions, listicles rose to 40.86%. Navigational/local prompts leaned toward product pages, category pages, homepages, and profiles. Transactional prompts moved even more strongly toward product and category pages.
In other words, AI citation behavior shifted with the task. The content that helps a model explain a market is not always the content that helps it compare vendors, locate a business, or support an action.
| Prompt intent | Leading content types | What the AI user is asking for | Primary GEO implication |
|---|---|---|---|
| Informational | Articles 45.48%; listicles 21.68%; how-to guides 9.21% | Explain, teach, define, or contextualize | Build strong first-party expertise and useful educational depth. |
| Commercial | Listicles 40.86%; category pages 12.42%; discussions 11.44% | Compare, shortlist, evaluate, or recommend | Build decision-oriented and third-party recommendation coverage. |
| Navigational / local | Product pages 21.95%; category pages 18.31%; homepages 13.56% | Find a known destination, provider, or local option | Make entity, location, service, and destination pages easy to discover. |
| Transactional | Product pages 24.88%; listicles 16.87%; category pages 14.97% | Buy, book, sign up, or take action | Strengthen product/service detail and conversion-ready page architecture. |
GEO implication #1: build an intent architecture before a content calendar
Traditional content planning often starts with a keyword list and then assigns a format. A more useful GEO workflow starts one step earlier: identify the job the buyer is asking the AI system to perform.
A prompt such as “What is enterprise SEO?” needs explanation. “Best enterprise SEO agencies for global brands” needs a consideration set. “LSEO enterprise SEO services” is navigational. “Book an enterprise SEO consultation” is much closer to an action.
Those prompts may belong to the same topic cluster, but the evidence required to answer them well is different. One article cannot realistically perform all four jobs.
Map prompts by intent first, then map the content and sources needed to support each intent.
This prevents a common failure mode: a brand with hundreds of educational pages assuming that content volume alone should translate into recommendation visibility.
GEO implication #2: owned content and third-party content solve different problems
The study reinforces why a modern GEO strategy cannot be built entirely on a brand's own domain. Owned content is essential for factual depth, expertise, product or service detail, technical accessibility, and conversion. But commercial recommendation prompts frequently pull from formats designed to compare multiple options.
That distinction becomes especially visible in professional services. In the study's deeper analysis of the top 1,000 professional-services URLs by citation count, 80.9% of the classified listicle citations came from third-party listicles, compared with 19.1% from self-promotional listicles. That does not mean 80.9% of all professional-services citations were third-party listicles. It means independent listicles dominated within the listicle subset the researchers classified.
The strategic lesson is not “buy more listicles.” It is that brands need credible evidence in the environments where buyers and AI systems evaluate alternatives. That may include Best Of lists, comparisons, reviews, expert roundups, buyer guides, directories, decision matrices, or other third-party formats depending on the query.
GEO implication #3: stop looking for one universal AI content format
The Wix/Peec research also undercuts the idea that there is one page template that can be “optimized for AI.” ChatGPT, Google AI Mode, and Perplexity showed some model-level differences, but the study concluded that intent was the stronger predictor.
ChatGPT had the highest article representation among the models studied. Google AI Mode showed a relatively balanced distribution across content types. Perplexity cited discussions at 17.35%, more than double the 7.52% cross-model average.
Those differences matter when a particular platform is commercially important to your audience. But for most brands, a strategy built around the buyer's task is more durable than publishing separate content simply because one model appears to prefer a certain format today.
“We need more AI content.”
This treats GEO as a new publishing category without defining which prompts, buyer stages, evidence gaps, or recommendation contexts the content must support.
“We need evidence for this buyer task.”
This starts with the prompt cluster, identifies the source types already shaping answers, and then builds or earns the missing evidence.
GEO implication #4: commercial visibility deserves its own strategy
Commercial prompts are where AI visibility becomes closely connected to buyer consideration. The user is no longer asking only what something is; they are asking which companies, products, or services deserve attention.
The 40.86% listicle citation share is significant because listicles naturally package multiple named options, criteria, rankings or groupings, and differentiation into a structure an AI system can use to assemble a shortlist. Discussions and category pages also become more prominent because they help contextualize choice.
For brands, that means commercial GEO should be audited separately from informational GEO. A company may be cited often for educational content while being almost invisible when the prompt changes to “best,” “top,” “recommended,” “for [use case],” “vs.,” or “alternatives.”
That is a recommendation gap, not necessarily an SEO content-volume gap.
GEO implication #5: conversion pages still matter
AI recommendation strategy can become overly focused on citations and third-party mentions. The study is a useful reminder that the buyer journey does not end at recommendation.
Product and category pages became far more prominent for navigational/local and transactional intent. That suggests brands still need clear service pages, product pages, category architecture, location information, pricing or offer details where appropriate, and strong conversion pathways.
This aligns with Google's current guidance that foundational SEO and useful, non-commodity content remain relevant to generative Search. GEO does not replace technical SEO, content quality, crawlability, or conversion architecture. It expands the visibility system around them.
What the study does not prove
Large citation datasets are useful, but they can be misused when descriptive findings are turned into universal prescriptions. There are several conclusions the research does not support.
Listicles cause AI recommendations
The study measured citation patterns. It did not run a controlled experiment showing that publishing or obtaining a listicle placement causes a brand to be recommended.
40.86% should become your content budget
Citation share is not an allocation formula. Your investment should reflect your own prompt mix, business model, competitive gaps, and existing content footprint.
One study predicts every market
The research covered specific models, industries, prompt construction, and a particular period. Your market can differ, so citation research should inform testing rather than replace it.
Third-party automatically means credible
Publisher relevance, evidence quality, methodology, freshness, disclosure, and editorial integrity still determine whether a placement is useful to buyers or defensible strategically.
Turn the study into a practical GEO operating model
The most useful way to apply the research is not to copy its citation percentages. Use it to build a repeatable system for identifying where your brand lacks the right type of evidence.
Build a prompt portfolio
Collect the questions buyers ask while learning, comparing, locating, validating, and acting. Include category prompts, “best for” prompts, comparison prompts, branded prompts, and conversion-oriented questions.
Classify each prompt by intent
Do not rely on keyword modifiers alone. Ask what job the user expects the AI system to perform: explain, shortlist, locate, or enable an action.
Audit the sources currently shaping answers
Record which publishers, pages, and content types appear. Look for recurring source categories rather than assuming your market matches the aggregate study.
Map evidence gaps
Separate owned-content gaps from third-party recommendation gaps. A missing service page requires a different response than absence from independent “best” or comparison content.
Build the right content type for the job
Create educational depth where buyers need explanation, recommendation-oriented coverage where they need comparison, and strong product/service pages where they need validation or action.
Measure visibility by intent cluster
Track mentions, citations, recommendation frequency, competitors, cited sources, clicks, and leads separately for informational and commercial prompts. A single blended score can hide the exact gap you need to solve.
Industry still matters—but it should refine the strategy, not lead it
The study did find meaningful vertical differences. SaaS had the strongest listicle bias at 35.37%. Professional services followed with listicles at 25.24%. Health and wellness leaned more heavily on articles. eCommerce was more balanced across listicles, articles, and category pages. Home repair showed a flatter distribution.
Those differences are useful as a second layer. Start with intent, then refine by vertical. A health brand answering a commercial query may still require stronger authority and evidence than a SaaS tool roundup. A professional-services firm may need more independent third-party recommendation coverage than a brand selling a commodity product with strong category-page demand.
The study therefore supports a two-dimensional planning model: buyer intent first, market context second.
What this means for LSEO's GEO framework
For LSEO, the research supports a broader definition of GEO than simply optimizing pages so an AI system can parse them. A modern program needs at least four connected layers:
- Search foundation: crawlable, technically sound, useful owned content.
- Entity and narrative clarity: consistent information about who the brand is, what it does, and where it fits.
- Third-party recommendation context: credible independent sources that place the brand inside relevant comparisons, Best Of lists, reviews, guides, roundups, and other decision formats.
- Measurement: prompt-level tracking of mentions, citations, recommendations, competitors, traffic, leads, and downstream outcomes.
This is also why Mention Engine is not positioned as a replacement for SEO or owned content. Its job is narrower and more specific: help brands systematically build third-party recommendation visibility around the content types and commercial questions where independent context matters.
The study is most valuable as a map of buyer tasks, not a scoreboard of content formats.
Articles, listicles, product pages, category pages, discussions, comparisons, and guides all have different jobs. GEO becomes stronger when the brand builds the right owned and third-party evidence for each stage instead of chasing whichever format has the biggest headline percentage.
Use citation research to decide where to act next
The practical question is not whether your brand should publish more articles or appear in more listicles. It is where the current AI recommendation footprint fails to match the buyer journey.
If your brand is absent from informational answers, strengthen expertise and educational depth. If it disappears during commercial evaluation, map the third-party sources and decision formats shaping the shortlist. If buyers can find you but cannot validate or convert, improve owned service, product, category, and conversion pages.
The Wix/Peec dataset gives marketers a valuable starting hypothesis. Your own prompt, source, competitor, traffic, and lead data should determine the final strategy.
Build the evidence AI needs at each stage of the buyer journey.
Mention Engine helps brands identify and build relevant third-party recommendation content around the commercial questions, formats, and publisher environments that influence AI consideration.
Explore Mention Engine