Case Study

The five most common AI visibility gaps we keep finding, golf clubs and local businesses

We've written individual posts on schema errors, llms.txt, NAP consistency, mobile speed and logged in bias as they've come up. Put together, a clearer picture emerges. Across every AI Recommendation Score test run completed so far, spanning golf clubs and a range of local service categories, the same handful of gaps keep reappearing, in different businesses that have never spoken to each other. That consistency is itself the useful finding: these aren't obscure edge cases, they're the default state for a business that has never checked what AI is actually saying about it.

1. No llms.txt file, anywhere

In every test run completed so far, the business being tested had no llms.txt file at the root of its domain. GPTBot, ClaudeBot, PerplexityBot and Google-Extended are almost always allowed to crawl freely in robots.txt, but the extra step of writing a short, plain language summary specifically for a model to read has simply not been taken up yet. It costs nothing, carries no downside, and remains one of the easiest, fastest fixes on any report we produce.

2. Schema markup that's broken, not missing

Missing schema is common. Broken schema is more common, and worse, because a model has no way to know the data is wrong. The recurring pattern is a business name sitting in the wrong field, an empty phone or email field despite the block otherwise existing, or an address reduced to a postcode with no street. A schema block like this actively feeds AI systems incorrect information rather than simply providing none, which is a worse starting position than having no structured data at all.

This is worth repeating because it runs against instinct. Business owners assume "we have schema markup" means the box is ticked. In practice, whether that markup is accurate matters more than whether it exists, and accuracy is the part almost nobody checks after the initial setup.

3. Directory and citation details that don't match

Small mismatches between how a business's name, address and phone number appear on its own site versus its Google Business Profile, Facebook page and directory listings show up constantly. A club listed as "St Example Golf Club" on its website and "St. Example Golf Club Ltd" on a directory looks trivial to a human. To a model cross referencing sources to build confidence in a fact, inconsistency reads as ambiguity, and ambiguity tends to get resolved by simply not mentioning the business rather than guessing which version is correct.

4. Mobile pages that load too slowly to be read

Across the sites we've tested, mobile PageSpeed scores well below what would be considered acceptable are common, not rare. A slow rendering page isn't just a poor visitor experience, it's a crawlability problem: if a crawler struggles to render key content in a reasonable time, that content effectively doesn't exist for indexing purposes, which means it can't be cited in an AI answer regardless of how good the underlying copy is.

5. Content that describes rather than answers

The most consistent gap of all isn't technical. It's that most business websites are written to describe a business in general terms, "a warm welcome," "excellent service," "a fantastic experience," rather than to answer the specific questions a prospective customer is actually asking. An AI model can't quote an adjective with confidence. It can quote a fact: a price, a room count, an opening time, a waiting list status. Every category we've tested rewards specific, citable detail over polished but vague marketing language, without exception so far.

GapHow common so farTypical fix effort
Missing llms.txtPresent in every test run to dateSame afternoon
Broken schema fieldsMore common than missing schema entirelySame afternoon to a few days
NAP/citation inconsistencyFound in the majority of businesses testedA few days of directory cleanup
Poor mobile page speedCommon, particularly on image heavy sitesDays to a few weeks depending on the cause
Vague, adjective led contentThe most consistent gap across every categoryOngoing, page by page

Why this pattern matters commercially

None of these five gaps are exotic or expensive to fix. That's the actual takeaway. The businesses we've tested aren't losing AI recommendations because of some deep structural disadvantage, they're losing them because nobody had checked, and a handful of specific, diagnosable issues had quietly accumulated. The businesses that test this early, while most of their category still hasn't, get to fix a short list of concrete problems rather than compete against a field that's already caught up.

Where to start

Check your own site against the five gaps above before assuming a bigger rebuild is needed. An llms.txt file, a schema validator pass, a directory listing audit, a mobile PageSpeed check, and a read through of your key pages asking "does this answer a real question or just describe us" will surface most of what's actually holding you back.

Find out which of these gaps apply to you.

We run the panel test across ChatGPT, Gemini, Claude and Perplexity for golf and local service businesses, then do the implementation work to fix what's holding you back.

Get your free AI Snapshot