You Rank on Google. So Why Won’t AI Recommend Your Firm? 15 Website Mistakes We Keep Finding

Something changed in how clients find professional-services firms, and it happened faster than most firms noticed. According to BrightLocal’s Local Consumer Review Survey 2026 AI supplement, 45% of consumers used AI tools to find local business recommendations in the past year – up from 6% the year before. One scope note on that figure: BrightLocal’s survey concerns local-business recommendations generally, not professional-services firms specifically – read it as a directional signal that AI-assisted discovery has gone mainstream, not as a sector-specific demand estimate. The question “which solicitor should I use?” is now routinely put to ChatGPT, Gemini, Perplexity and Google’s AI Overviews before it is ever put to a colleague.

Where do those engines get their answers? Mostly from you. When BrightLocal analysed where AI engines source their local business information in its July 2025 study of AI search sources, it found that “the vast majority of sources across every single LLM and industry were businesses’ own websites”. The platform-by-platform detail behind that finding lives in our companion piece on what AI actually reads on your website.

That reframes what your website is for. It is no longer a brochure that prospects may or may not read; it is the interview your firm sits every time an AI engine considers whether to recommend you. And in the website audits we run for law firms, financial-services businesses and consultancies, we keep finding the same fixable mistakes – the answers, so to speak, that fail the interview.

One scope note before the list. This article covers the website itself – the artefact AI engines actually read. For the practice side of the discipline, the 12 SEO mistakes professional-services firms are still making is the companion piece. If you want the sector-specific view, we cover digital marketing for lawyers, financial services and consultancies separately.

Here is what we actually find.

The broken basics

Every audit starts with the unglamorous checks, and the unglamorous checks keep paying for themselves. These are not subtle optimisation opportunities. They are defects a prospective client can spot in thirty seconds – and that an AI engine ingests as evidence about how carefully the firm is run.

1. Forms that don’t work

The single most expensive mistake we find, and we find it more often than anyone would guess. A firm we audited had a contact page whose enquiry form displayed as a line of raw shortcode – the plugin behind it had been deactivated at some point, and the text

Error: Contact form not found.

had been sitting where the form should be, in front of every visitor, for months. Nobody inside the firm had noticed, because nobody inside the firm ever visits their own contact page.

The quieter variants are worse, because they look fine. Forms that submit to a mailbox nobody monitors. Forms whose notification emails land in spam, so enquiries arrive – and die – in a folder no one opens. In each case the firm was paying to generate enquiries that a broken final step silently discarded. For a machine reading the page, a form rendered as shortcode text is also a legibility problem: the page’s primary conversion element is literally gibberish.

Self-check: submit your own contact form today, from a personal address – and put a recurring reminder in the diary to do it monthly.

2. Placeholder and dead links

Social icons in the footer that link to “#” – the placeholder the theme shipped with, never wired up to a real profile. Icons that point at the platform’s homepage rather than the firm’s page. Footer navigation still linking to pages that were removed two redesigns ago. We have found all three on otherwise polished sites, sometimes on the same site.

Individually these look trivial. Collectively they are a signal – to a visitor, that details are not this firm’s strength; to a crawler, that the site’s own claims about itself cannot be followed and confirmed. Links are how machines verify: an AI engine that tries to confirm your LinkedIn presence and lands on “#” learns precisely nothing, at the moment it was ready to learn something in your favour.

Self-check: click every link in your own footer and header, including every social icon. It takes four minutes.

3. Test pages leaking into the index

Run this query in Google: site:yourfirm.co.uk. It lists every page of yours Google has indexed – and for a surprising number of firms, that list includes staging pages, template demos, and pages titled some variant of “Test – DO NOT PUBLISH”. We keep finding them not just indexed but listed in the site’s own sitemap.xml, formally invited into the index, with robots directives doing nothing to stop them.

In the AI era this failure has a new cost. Your sitemap contributes to discovery – it tells search engines which URLs to crawl – and any unwanted page that ends up indexed becomes part of the web corpus available to search engines and to the AI systems built on top of them. A test page indexed alongside your service pages is a credibility wound you never see happen: it sits in the public record of your firm, saying, in effect, “we do not control our own website”. We have written up how this happens and how to close it in our field guide to the leaking sitemap, and it is one of the first items in the crawl section of our SEO & Marketing Intelligence Report.

Self-check: run the site: query and read your sitemap.xml this week. Flag anything you would not put in front of a client.

4. CTAs that disappear

The homepage says “Book a consultation” in three places. The service pages – the pages people actually land on from search, and the pages AI engines quote when asked what you do – say nothing at all. We see this pattern constantly: conversion thinking was applied to the homepage during the last redesign and never propagated to the pages that receive the traffic.

Sometimes the problem is inconsistency rather than absence: one page asks visitors to call, another to email, a third to download a brochure. A visitor who has to work out how you want to be contacted usually will not. The same ambiguity degrades how clearly a machine can describe your next step to someone asking “how do I get in touch with this firm?”.

Self-check: open your three busiest service pages as a stranger would. Is there one clear, consistent next step on each – without scrolling back to the homepage?

Technical trust minimums

None of what follows is news, and that is rather the point: these are the minimums, and audits keep showing firms below them. We have covered each in depth elsewhere, so this section stays brief and links out.

5. Slow on mobile

Speed stopped being a nicety when Google formalised Core Web Vitals – LCP, INP and CLS – as measurable thresholds, and it has become a trust signal in its own right: a site that takes six seconds to load on a phone loses a proportion of its visitors before showing them anything, and tells every measuring system that the experience behind the brand is poor. We have explained what Core Web Vitals actually measure and what a slow website really costs a firm in dedicated guides; the short version is that this is the most quantified mistake on this list.

Self-check: run your busiest service page – not your homepage – through PageSpeed Insights and look at the mobile score first.

6. HTTPS and mixed-content faults

A padlock in the address bar is the least a professional firm’s website can offer, yet we still find certificates that have lapsed, internal links pointing at the http:// versions of pages, and mixed-content warnings triggered by images loaded insecurely into secure pages. To be clear about where this sits: a secure, cleanly served site is a prerequisite for reliable crawling and user trust – not an advanced GEO tactic. Browsers warn visitors away from mixed content, and a lapsed certificate can turn every page on the site into a warning screen overnight. This is table stakes, and audits keep finding firms below it.

Self-check: open your key pages and look for the padlock; then view any page over http:// and confirm it redirects to https://.

7. Crawl-control confusion

Two failure modes, opposite in direction. The first is the robots.txt accident – a Disallow: / left over from a development environment, quietly instructing every crawler to ignore the site. The second is having no considered crawler policy at all: nobody has decided which crawlers – search or AI – may read what, so the site’s access rules are whatever a default left behind. robots.txt remains the operative control here, and it deserves a deliberate review; llms.txt is an emerging, optional convention some site owners are adopting alongside it. Neither is a guaranteed AI-visibility control – the mistake is not the absence of any one file, but the absence of a decision. Our guide to llms.txt and AI search readiness covers the sensible middle.

Self-check: read yourdomain.com/robots.txt – all of it – and confirm every rule in it is one you actually chose; then decide, deliberately, whether an llms.txt is worth adding.

What AI reads – and can’t find

If Category 1 was about defects, this category is about absence. AI search systems don’t all retrieve or evaluate websites the same way – but across the platforms studied, the same foundational website signals repeatedly matter. The BrightLocal findings quoted above mean those systems arrive at your website expecting to find the facts of your firm, and the mistakes below are the facts most often missing – the questions your site is asked and cannot answer. (For the per-engine detail on what each platform retrieves and cites, see what AI actually reads on your website.)

8. Your name, address and phone number are nowhere on the site

It sounds impossible for a professional firm, but we find it: sites where the office address appears only inside a Google Map embed (invisible to a text reader), where the phone number exists only as an image, or where the details on the contact page disagree with the firm’s Google Business Profile and directory listings. Machines cross-reference; disagreement between your website and your listings weakens the one entity signal that should be effortless. Name, address and phone – consistent to the character with your Google Business Profile – belong in crawlable text on the homepage and the contact page at minimum.

Self-check: select-and-copy your address and phone number from your homepage. If you cannot select them as text, a crawler cannot read them.

9. One vague “Services” page instead of one page per service

BrightLocal’s research into how AI engines answer service queries points at a simple test: when someone asks “does this firm handle commercial leases?”, the engine looks for a page that says so. A single “Our Services” page listing twelve practice areas in one paragraph each fails that test twelve times – there is no page about the service, so there is nothing to retrieve, cite or quote. This is the mistake with the clearest structural fix: one properly written page per service you actually want to be found for.

Self-check: pick a service that matters to your revenue and ask an AI engine, “does [your firm] offer [that service]?” If the answer is vague, so is your site.

10. No question-answering content

AI engines are asked questions, and they quote whoever answers them. A site with no FAQs, no plainly phrased answers, and no content addressing the questions clients actually ask (“how long does probate take?”, “what does an audit cost?”) gives the engines nothing quotable – so they quote a competitor, or a directory, instead. Answering real questions in direct prose is the single most reliable way to earn a presence in AI answers, and it is the core of our GEO-ready website checklist.

Self-check: list the five questions every new client asks in the first meeting. Does your website answer any of them in plain text?

11. Missing structured data

Schema markup – Organization, LocalBusiness, Service, FAQPage – is the layer of your site written specifically for machines: it states, unambiguously, what kind of entity you are, what you offer and how you can be reached. Sites without it are asking every crawler to infer from prose what could have been declared in data. Most professional-services sites we audit have either no structured data or a half-configured default emitted by their theme. Two caveats keep this honest: markup must match what the page visibly says – invalid or misleading markup may simply be ignored – and schema guarantees nothing: no rich result, citation or recommendation follows automatically from adding it. Use the types appropriate to the entity (Organization, Person, Service, Article) and let them declare what the page already demonstrates. The full picture is in our guide to structured data for AI search.

Self-check: paste your homepage URL into Google’s Rich Results Test. “No items detected” means you are making the machines guess.

12. Weak entity signals – the ghost-citation problem

This is the mistake behind this article’s title. Semrush’s Ghost Citations study (June 2026) found that most AI citations never mention the brand by name: the engine uses a firm’s content to build its answer, links the source, and names the firm not at all. Your expertise informs the recommendation – and someone else’s name is on it.

Ghost citations are usually an entity problem, not a content problem: the content was good enough to use, but the brand was not legible enough to credit. The fixes – naming conventions, entity-consistent site architecture, and the trust signals AI engines look for in regulated sectors – are covered in depth in our companion piece on the ghost-citation problem.

Self-check: ask ChatGPT and Perplexity what they know about your firm by name. Thin, generic or outdated answers mean weak entity signals.

Credibility AI can verify

AI engines have no reason to take your word for anything, and across platforms the recurring pattern is triangulation – your site against your reviews, your claims against your named people. Two mistakes make that triangulation fail.

13. Reviews nowhere in sight

The Trustpilot and Seer Interactive study of more than 800,000 AI answers found that actively-reviewed brands were cited in 75.3% of relevant AI answers, against roughly 1% for brands without active review presence. That is not a gap; it is a wall. Yet we audit professional-services sites with no visible testimonials, no link to any review platform, and no connection between the website and the firm’s Google Business Profile reviews. If reviews exist but your site never surfaces or links them, you are hiding what the available research suggests is among the strongest publicly observed reputation signals – though the association is observational: heavily reviewed brands differ from unreviewed ones in many ways, and no study has isolated reviews as the cause.

Self-check: can a visitor – or a crawler – get from your homepage to a live, third-party review profile in one click?

14. A faceless firm

An “About” page of stock photography and abstract values, no named partners, no photographs of real people, no biography a machine could connect to a directory entry or a LinkedIn profile. BrightLocal’s guidance notes that Google’s own teams have confirmed AI-generated results reward first-hand content – the detail that only someone who has actually done the work could write. A faceless site offers none of it: no people to verify, no experience to detect, no reason to prefer your firm’s account over a directory’s. Professional networks matter here too – LinkedIn is among the most-cited domains in AI answers to professional-services queries, and your named people are the bridge between your site and their profiles there.

Self-check: does your About page name real people, with real photographs, whose profiles exist somewhere off your site?

Human-readability is machine-readability

The final category is one mistake and a pattern worth naming: everything that makes a site harder for a person to read makes it harder for a machine to parse.

15. Accessibility and legibility failures

Grey 11px text on white. Headings chosen for their font size rather than their hierarchy – an h4 here, a styled paragraph there, no h1 anywhere. Contrast that fails WCAG thresholds. These are usually filed under “design polish”, but they are structural: semantic HTML – a genuine heading hierarchy, properly marked-up lists, real landmarks – improves accessibility, browser interpretation and structured extraction alike. The same properties that let a screen reader navigate your page cleanly are the ones that make its content easier for automated systems to interpret – which is why accessibility is a first-class concern in our web design practice, not a compliance afterthought.

Self-check: run one key page through a contrast checker, and view your page’s heading outline (any browser accessibility inspector shows it). If the outline reads as nonsense, so does your page.

Two further habits belong here at mention level. Stale content – a “News” page whose latest entry is three years old – actively testifies against you, to humans and machines alike, and quietly dates every claim on the site. And self-focused copy – pages about how excellent the firm is rather than about the client’s problem – fails both readers the same way: it answers no question anyone asked. Both tend to be symptoms of the same root cause: nobody owns the website after launch.

The pattern behind all fifteen

Look back down the list and one thing should stand out: not one of these mistakes required new technology to fix, and not one of them is really an “AI mistake”. A working form, honest links, a clean index, clear services, real people, visible reviews, legible pages – these are exactly what a careful human visitor has always wanted. What has changed is that the visitor is now sometimes a machine with a perfect memory and a habit of quoting – and it reads everything, including the pages you forgot you had.

That is the practical meaning of the research this article opened with. AI engines lean on your website more than on any other source; therefore the trust signals you show humans and the trust signals you show machines are the same signals. Firms that treat AI visibility as a bolt-on tactic miss this. Firms that fix the fifteen items above are doing generative engine optimisation whether they call it that or not – and the effect is measurable, month by month, as an AI Visibility Score.

The five mistakes we’d fix first

Fifteen items is a programme, not a to-do list. If you can only fix five, fix these, in this order:

Priority	Mistake	Why first
1	Indexed junk and test pages (mistake 3)	Reputation and index hygiene: everything indexed under your domain is part of your public record, and cleaning it costs almost nothing.
2	Broken conversion paths (mistakes 1 and 4)	Direct revenue: every enquiry a broken form or missing CTA discards was already paid for.
3	Missing service pages (mistake 9)	Search and AI discoverability: without a page per service, there is nothing to rank, retrieve or cite.
4	Weak entity signals (mistake 12)	Brand attribution: this is what decides whether your content is credited to you or quoted namelessly.
5	Missing first-hand expertise (mistake 14)	Trust and differentiation: named people and lived detail are the evidence no directory or competitor can copy.

Frequently asked questions

Where to start

Every mistake in this article is checkable, and most of the self-checks take minutes – so start there, this week, with the site: query and your own contact form. If you would rather have the first pass run for you, start with our free website SEO analysis – it covers the technical and on-page basics automatically.

The firms we audit are rarely making exotic mistakes. They are making these fifteen – and the ones that fix them are the ones the engines, human and artificial alike, end up recommending.