TL;DR
I have now audited 404 websites for AI visibility across 176 unique domains, with 67 of those audits running at high-reliability standards and 103 sites carrying repeat audits over time. The data is not flattering for most of financial services. Twenty-three percent of audited sites score below 50 on AI/GEO discoverability, which is a polite way of saying they are nearly invisible to AI answer engines. Forty-four percent score below 60 overall. Accessibility is the single biggest gap across the dataset, sitting at an average score of 50. Below are the four numbers worth committing to memory, the eight fixes that keep showing up, and what the high-scoring sites are doing differently.
I want to lead with the methodology before the numbers, because the methodology matters. Each audit evaluates a website across five categories. GEO/AEO for AI discoverability. SEO for search visibility. Technical for site performance. Accessibility for ADA and WCAG conformance. Reputation for external signals. A site gets a score in each category and an overall score. Some sites have been audited once. Some have been audited several times across months, which lets me track movement. Sixty-seven of the audits ran with full high-reliability proxies, which means the responses are independently verified across rendering, location, and device. The rest are still useful, but the high-reliability subset is the cleanest signal.
Here is what the data says.
The four numbers worth committing to memory
Twenty-three percent of audited sites score below 50 on AI/GEO discoverability. That category measures whether an AI tool can confidently identify, summarize, and cite the site when asked a relevant question. A score below 50 is functionally invisible. Almost a quarter of the sites I have audited are missing from the AI surface entirely. This is not a fringe problem.
Seventeen percent of financial institution websites pass the high-reliability audit threshold. That is sixty-seven of four hundred and four. The other eighty-three percent are either falling short of the bar or have not yet been re-audited at the higher standard. Either way, the share of sites operating at the level the AI surface actually rewards is small.
Fifty is the average accessibility score across the entire dataset. Accessibility is the single biggest gap I see across all the categories. It is not because accessibility is the most important category. It is because it is the most consistently neglected. The same five fixes show up across the failing sites. None of them are exotic. All of them are documented in the WCAG 2.1 guidelines.
Forty-four percent of audited sites score below 60 overall. Sixty is the rough line where a site goes from “needs work” to “good enough to compete.” Almost half of the sites I have audited are below that line. Most of them are reachable inside a quarter with focused product, content, and engineering work.
Accessibility is the number one gap, and not for the reason you think
Forty-five percent of audited sites fail the accessibility category outright. The percentage failing in other categories is meaningfully lower. SEO at 37 percent failing. Reputation at 29 percent. GEO/AEO at 23 percent. Technical at 16 percent. Accessibility is the outlier.
The interesting part is why. Accessibility is not the hardest category to improve. The fixes are well-known. ARIA labels, form labels, viewport meta tags, alt text, focus states. None of these are new. None of them require a new vendor. Most of them can be addressed by a competent front-end engineer in a couple of sprints.
The reason accessibility is the biggest gap is that it is the easiest category to deprioritize. It does not show up in revenue dashboards. It does not get flagged by AI tools the way structured data does. It quietly degrades the experience of a meaningful share of your audience without producing a single line item in a report. That makes it the first thing pushed off the backlog every quarter.
The cost is real. AI tools that surface site previews, that read alt text, that parse form fields, and that depend on semantic markup are penalizing accessibility failures whether the team realizes it or not. Accessibility is no longer a compliance and equity story. It is also a discoverability story.
The eight fixes that keep showing up
Across all 404 audits, eight specific fixes are prescribed more often than any others. The list is worth posting next to the engineering team’s quarterly planning board.
Implement structured data appears in 83 of the 404 audits. Structured data is the single biggest GEO gap in the dataset. AI tools depend on it to understand the entity, the products, and the offerings on a page. Sites without it are forcing AI to infer, and AI does not infer reliably.
Implement lazy loading appears in 50 audits. Lazy loading is a technical performance fix. Sites that do not implement it are penalized by both human and AI rendering, and the fix is well understood.
Implement ARIA labels appears in 37 audits. Add ARIA labels appears in another 35. Combined, that is 72 sites where the basic semantic layer for accessibility is missing.
Claim and optimize review profiles appears in 35 audits. This is a reputation fix. Sites that have not claimed their profiles on the major review platforms are leaving reputation signals on the table that AI tools weight heavily.
Add viewport meta tag appears in 27 audits. This is one of the most basic mobile-readiness fixes, and a quarter of the way through the dataset, it is still missing on a meaningful share of sites.
Add form labels appears in 23 audits. Form labels are required for both accessibility and AI parseability of conversion paths.
Set canonical URLs appears in 21 audits. Sites without canonical URLs are confusing both search engines and AI tools about which version of a page is authoritative.
None of these eight fixes are dramatic. All of them are overdue.
What the high-scoring sites are doing differently
The top of the leaderboard is interesting. BECU, Summit Credit Union, Wells Fargo, Bank of America, American Express, and a small group of others sit at 85 overall. Below them at 82 are Alliant Credit Union, GTE Financial, Navy Federal, Numerica Credit Union, Suncoast Credit Union, and TDECU. These sites are not all the largest in the sample. They are not all the ones with the biggest marketing budgets. What they share is consistency across all five categories.
A site that scores 85 overall is not winning by being world-class in one category. It is winning by hitting the 75-to-90 range in each of the five. The pattern is balance. Strong GEO, strong SEO, strong technical, strong accessibility, strong reputation. No single category is doing the heavy lifting. The composite is what wins.
The lesson, especially for financial institutions trying to make a quarterly improvement, is that the path to the top of the leaderboard is not through one heroic project. It is through closing four or five medium-sized gaps in parallel.
What the data does not tell me
The data does not tell me whether a higher visibility score correlates with deposit growth, member acquisition, or revenue. I believe it does, because the underlying mechanics suggest it should, but the audit measures the surface, not the outcome.
The data also does not tell me which AI tools the prospects of any given institution are using most. The audit measures whether a site is discoverable across the major AI tools. The actual usage mix by audience varies.
These are real limits. They are also second-order. You cannot study the outcome until you can measure the surface. The audit is the measurement instrument. The harder questions follow.
The bottom line
Four hundred and four audits across 176 sites is not the whole financial services market. It is enough to see the pattern. The pattern is that a meaningful share of financial institutions are nearly invisible on the AI surface, that accessibility is the most neglected category by a wide margin, that the same eight fixes show up across the failing sites, and that the path to the top of the leaderboard is balance, not heroics.
If you have not run an audit on your own site, I would do that this quarter. The methodology is open. The findings are reproducible. The fixes are tractable.
The institutions that move now will look meaningfully different in next year’s data than the ones who do not.
Kevin Farley writes about AI visibility, AI readiness, and strategic growth for financial services. Read more on the blog.