Short answer: Fixing your site templates beats buying a GEO tool. SISTRIX analyzed 96,418 URLs and found pages with a table of contents were 1.7x as likely to be cited by AI, Article schema 1.6x and tables with header rows 1.5x, while llms.txt showed no measurable effect.
On October 22, German SEO data company SISTRIX published its "citation worthiness" study, drawing about 16,000 cited URLs each from ChatGPT, Google AI Mode and AI Overviews, for 96,418 URLs across 17,380 domains. What stands out is the design: each cited page was paired with a non-cited page from the same domain, removing brand and domain-strength effects to isolate the page itself. Pages meeting most structural features were cited 2.1 times as often as barely optimized pages.
Why does this matter now? GEO (generative engine optimization) became marketing's hot topic over the past year, with endless "get AI to recommend you" tactics, yet most studies use a few hundred URLs. SISTRIX tested one feature at different sample sizes: at 833 URLs, FAQ schema appeared to lower citations by 23%; at 4,600 URLs it raised them by 31%; only at 96,418 URLs did it settle at a 19% lift. In other words, GEO studies with fewer than about 5,000 URLs can point in the opposite direction. The same week, SparkToro's Rand Fishkin warned that several studies show AI still drives under 1% of traffic, so budgets should not go all-in on AI visibility.
The tool market is consolidating. Moz announced a data partnership with Profound to put rankings and AI mentions in one report; Microsoft Clarity added AI visibility to its free dashboard; Semrush and Ahrefs both launched AI tracking modules. These tools measure; the SISTRIX study answers what to change. Google is shifting too: Search Engine Roundtable reported on October 23 that citation cards beside AI Overviews now show AI-rewritten summaries, so your own description may not be used verbatim, making clear page structure even more important.
For Taiwan SMBs, the takeaway is simple: before paying monthly to track whether AI mentions you, make sure your templates are machine-readable. The good news is that fixing a template once improves every article on the site. Below: the numbers, actions for three readers, tools compared and a no-subscription alternative.
The details: structure matters most, trust signals second, llms.txt and HowTo make no difference
The study boils down to one line: "structure for the machine, trust for the human."
| Page feature | Citation odds factor | Note |
|---|---|---|
| Table of contents | 1.7 | Only 14% of pages have one |
| Article schema / at least five H2s | 1.6 | Strongest on brand websites |
| FAQPage schema / table with header row | 1.5 | Tables reach 1.8 on ChatGPT |
| Author box with photo | 1.4 | Top trust signal |
| TL;DR at top / visible updated date | 1.3 | Only 5% show an updated date |
| llms.txt, HowTo, Breadcrumb schema | ~1.0 | No measurable difference |
| Product schema | 0.8 | Just 0.5 in AI Overviews |
Platforms differ: ChatGPT favors question titles (1.9x) and question H2s; Google AI Mode is an all-rounder, with table of contents and five-plus H2s at 1.8x each; AI Overviews is the pickiest, with Article schema at 2.0x. Limitations: DACH region only, correlation not causation, and Perplexity and Claude were not tested.
Actions for three readers: change the template and fix the whole site at once
The most efficient move is to change templates, not rewrite articles one by one.
- Brand owners: open your 5 most important service or knowledge pages and check for a table of contents, tables, FAQ and author bio; if 3 or more are missing, schedule it for next quarter's update.
- Marketers: open new articles with a 2–3 sentence direct answer, turn H2s into questions readers ask, and put comparisons in tables; e-commerce teams should invest in buying guides, not product pages.
- Developers: make article templates auto-generate the table of contents, Article and FAQPage JSON-LD, author box and updated date, and drop inflated hopes for llms.txt.
AI visibility tools compared: trackers measure outcomes but cannot fix pages
These tools tell you whether you were mentioned; none will rewrite your template.
| Tool | Positioning | Pricing | Best for |
|---|---|---|---|
| SISTRIX AI Index | Large-sample AI citation data, DACH and Europe focus | Paid subscription | Teams in European markets tracking industry trends |
| Profound | Dedicated tracking of brand presence in AI answers | Paid, enterprise-leaning | Mid-to-large brands with budget |
| Semrush / Ahrefs AI modules | AI tracking added to existing SEO suites | Requires a base plan | Teams already on those suites |
| Microsoft Clarity | Heatmaps plus AI visibility and referral reports | Free | Budget-limited SMBs |
What they won't tell you: correlation is not causation, and the "current year" is the biggest trap
The best use of this study is to set priorities, not to copy every item.
- The 2.7x year-in-title effect is an illusion: the authors say it reflects maintained content. Changing the year on an old article's title likely does nothing and may look manipulative.
- Length is not the cause: 1,000-word pages were cited more mainly because they were better structured. Padding dilutes the point.
- Citations are not sales: Google is rewriting summaries in citation cards, so a citation may not bring a click. Tracking conversions still matters more than tracking mentions.
A no-subscription alternative: one template checklist plus free tools
A single checklist covers 80% of the work for most SMBs.
- ☐ Templates auto-generate a table of contents and at least five H2s
- ☐ Every article has Article and FAQPage JSON-LD, validated with Google's Rich Results Test
- ☐ Comparisons, prices and specs always use HTML tables with header rows
- ☐ Add an author box (photo + bio) and a visible updated date
- ☐ Review AI referral traffic and conversions monthly in Microsoft Clarity and GA4
FAQ
Do we need an llms.txt file?
There is no evidence it helps yet. In the SISTRIX study, 11.7% of cited URLs came from domains with llms.txt, below the 14.0% share across all domains.
Is FAQ schema still worth adding?
Yes, if it reflects real questions. In the large sample, pages with FAQPage were about 1.5x as likely to be cited; padding irrelevant questions does not help.
How do e-commerce product pages get cited by AI?
Do not rely on product pages. They were cited half as often in AI Overviews; write buying guides with Article schema instead.
Are longer articles cited more often?
No. The 1,000-word effect mostly reflects better structure. Fix heading hierarchy and tables before adding words.
Should we buy an AI visibility tool?
Not yet for low-traffic SMBs. Use free Microsoft Clarity and GA4 to see AI referrals, then reassess after fixing templates.
My take: GEO is not a new skill but website basics left undone for years
The mainstream claim is that AI search requires an entirely new GEO strategy. My view is the opposite: the tactics that work here, tables of contents, heading hierarchy, tables, structured data, author information, are SEO basics from a decade ago that most sites never got right. Only 10% of cited pages already met most features, so the barrier is not technology but someone willing to fix the template. Over the next 12 months, services that charge for "AI visibility reports" will face churn as clients realize reports change nothing.
For ScriptWalker (a Taiwan Laravel/Flutter custom development studio), the service opportunity is an "AI-citable template upgrade": add auto tables of contents, JSON-LD, author boxes, updated dates and table styles to the Laravel article template once, and hundreds of existing articles benefit without rewrites. Clients are not buying a GEO buzzword; they are buying site structure fixed once that keeps working.
Sources
- SISTRIX: Citation worthiness study (2026-10-22)
- Google Search Central: AI features and your website
- Google Search Central: Article structured data
- llms.txt proposal
- SparkToro: Is AI Visibility a Marketing Boondoggle? (2026-10-22)
- Search Engine Roundtable: Google AI Overview Citation Cards With AI-Generated Summary Snippets (2026-10-23)