AI Search Readiness: We Tested 500 Hosting Sites

Last week I crawled the top 500 hosting company websites in our directory and measured every signal that determines whether an LLM (ChatGPT, Perplexity, Claude, Google AI Mode) is likely to cite a page when someone asks "what's the best hosting for X."

Key takeaway: Most hosting sites lack basic AI-search signals like FAQPage schema and llms.txt, leaving citations open for competitors to claim.

The numbers are worse than I expected.

Of the 437 sites that were reachable, zero scored 6 or 7 out of 7 on a basic AI-readiness checklist. Only four scored 5. The median host hits 2.

To put that in context: this is the most-trafficked category-defining industry of the modern web, and almost nobody is ready for the way buyers are starting to actually shop.

What I measured

Seven signals, each binary, each well-documented as a citation lever in published Generative Engine Optimization (GEO) research:

1. llms.txt at the root domain. The Jeremy Howard proposal from September 2024 that gives LLMs a structured map of your site, written in their grammar.
2. JSON-LD structured data anywhere on the homepage. The single biggest signal AI engines use to extract facts.
3. FAQPage schema specifically. Every paper on GEO since 2024 ranks this as a top-3 citation driver.
4. Organization schema with founding-date, sameAs, knowsAbout. The "who are you" signal.
5. BreadcrumbList schema. Helps AI engines understand site hierarchy without re-crawling everything.
6. SpeakableSpecification schema. The voice-and-AI surface that lets engines quote your passages verbatim.
7. Named statistics in body copy. Specific numbers ("99.9% uptime", "4.8/5 from 12,000 reviews", "20 years hosting"). The most-cited research on Princeton's GEO study showed this single thing lifts citation rate by 30 to 40%.

All seven are public, free, and shippable in a weekend. None require buying anything from anyone.

The headline numbers

From a sample of 437 reachable sites in the hosting industry, ordered by HostScore (so we are sampling the better end of the long tail, not random small operators):

A few of these deserve their own sentence.

Only 4.8% of hosting sites have FAQPage schema. This is the lowest-effort, highest-yield AI signal you can ship. Most hosts already have an FAQ on their pricing page. Wrapping it in <script type="application/ld+json"> takes 20 minutes. Ninety-five out of 100 of them have not done it.

Only 25% have llms.txt. The file was proposed nearly two years ago, and a quarter of the most engaged operators have shipped one. Three-quarters have not. The file is ten lines of plain text. There is no excuse.

Zero hosts have DefinedTerm schema. This one is harder to fault people for because it is less well-known, but it is the single most underrated AI-citation surface in 2026. If you have a proprietary metric (HostScore, your own uptime measurement, a trust signal, a benchmark), DefinedTerm is what makes ChatGPT and Perplexity cite you as the source for that term's definition. We use it at HostList for HostScore. The fact that not a single other host uses it for anything is striking.

What hosts are doing wrong, specifically

The pattern across the bottom-scoring sites was monotonous and easy to spot.

Marketing copy that names no numbers. "Industry-leading uptime", "blazing fast servers", "world-class support". LLMs cannot use these phrases. They cite specifics. A page that says "99.987% measured uptime in 2026 (source: report)" gets cited; a page that says "the most reliable hosting" does not.

FAQs as accordion HTML with zero schema. The data is there. It is just invisible to AI extractors. The fix is twenty lines of JSON-LD and an hour of QA. The lift is, by GEO research, 30 to 40% on cited-passage rate.

Marketing-page sprawl with no methodology page. Half the sites I looked at have ten pages about how great they are and zero pages explaining how a buyer would compare them. LLMs love methodology pages. They love published rubrics. They love anything that gives the engine a fact pattern to extract instead of an opinion to discount.

Robots.txt files that block GPTBot or PerplexityBot. 6.6% of the sample is actively blocking AI crawlers. I understand the impulse, but in 2026 that is the same as opting out of the next decade of how buyers will find you. If you do not want your content used to train models, fine; block the training bots. But blocking the search-time crawlers (ChatGPT-User, PerplexityBot) is opting out of citations. About 1 in 15 of the sampled hosts are doing this without realising it.

What good looks like

Five sites in the sample scored 4 or 5 out of 7. None of them are tier-1 hosting brands. The leaderboard:

A counter-intuitive finding sits in this table. The most AI-ready sites in the sample are not the household-name brands; they are smaller operators (often web-design and hosting-adjacent agencies) whose SEO setup is more modern. Brand-name hosts like WP Engine, Kinsta, and Bluehost score in the 2 to 3 range despite spending an order of magnitude more on marketing.

This is the same dynamic that played out in 2008 when the household-name hosts lagged on responsive design behind smaller operators. The big brands will catch up; the operators who are ready first will earn the citations during the transition window. That window is open now.

The seven-point checklist, in priority order

If you operate a hosting company, here is what to ship, in the order I would ship it:

1. FAQPage schema on your pricing page. Highest yield. Wrap the FAQ you already have in JSON-LD. Twenty minutes of work. Ship today.
2. Named statistics in your homepage hero. Replace "fastest hosting" with "measured 187ms median TTFB in 2026 across 12 global locations." Two hours.
3. Organization schema with sameAs. Declare who you are, where you are, when you started, and link your verified social profiles. One hour.
4. llms.txt at the root domain. Ten lines of plaintext. The header, two or three of your most important URLs, your one-line positioning, and a link to a methodology page. Thirty minutes.
5. BreadcrumbList schema on every category and product page. Helps AI engines navigate your site. One hour with a template.
6. A methodology page. Pick one. Uptime methodology, support-response methodology, pricing-fairness methodology. Document it. Cite the data sources. This is what positions you as authoritative when LLMs are sampling.
7. DefinedTerm schema for one proprietary metric. If you do not have one yet, this is the time. A 12-month uptime score, a measured benchmark, a published index. The first host in your sub-category to do this owns the AI-citation slot for that term forever, or until a competitor unseats you.

That whole list is one engineer-week of work. The win is multi-year.

What we ship at HostList (transparency)

In the interest of practising what I am preaching: every host profile on HostList ships all seven of these signals. That is not a flex; it is how we earn our own citations in Perplexity and ChatGPT and AI Overviews. The methodology is at /hostscore. The DefinedTerm for HostScore is on every page. The FAQPage and BreadcrumbList and Organization are auto-generated from the directory data. The named statistics are in the citation-lead passage that anchors every host page.

We rebuilt the whole site to do this in early 2026 because we believed (and still believe) that AI search is where hosting buyers will increasingly start their research. The first eight months of citations data have made it look right.

Methodology

I crawled 500 hosting company websites between 3 June 2026 and 4 June 2026, sampled from the top of our directory by HostScore (so we are testing the upper end of the industry, not the long tail of dormant Wordpress sites). 437 were reachable on the first attempt. Each site was probed for three resources (homepage HTML, /robots.txt, /llms.txt) and the homepage was scanned for the seven structured-data signals listed above. Source code for the audit is open: scripts/ai-readiness-crawl.ts in our GitHub repo.

Replication instructions, the full results JSON, and the per-host scorecard are all available; email me at [email protected] if you want the raw file.

The numbers are likely to look better in the next sweep, because once people realise where the bar is, the bar moves. That is fine. If this post nudges thirty hosting companies to ship FAQPage schema this month, the post has paid for itself.