This is a legitimate worry. I’ve been on the end of numerous phone calls and emails about it — and the pattern is consistent.
Most businesses don’t have an AI problem. They have one specific thing broken — and once you know what it is, fixing it is more straightforward than you might think. The challenge is that most people are fixing the wrong thing. They’re rewriting content when the problem is entity data. They’re chasing Google when the problem is Bing. They’re building links when AI systems already know them — they just don’t trust them enough to name them. Most businesses do not have an AI visibility problem everywhere. They have a bottleneck at one or two layers — and they are usually fixing the wrong one.
This guide maps the five layers where AI visibility breaks, gives you the diagnostic questions to find your specific gap, and tells you what to do about it in order of leverage. If you want the context and commercial framing first, the companion article is at Worried AI is killing your business enquiries?
This five-layer action plan was developed by Sean Mullins, Founder of SEO Strategy Ltd, from diagnostic work across client engagements in 2026. Each layer maps to a specific failure mode in the AI Discovery Stack. If you want a diagnosis of which layer is failing for your specific business rather than working through the checklist yourself, the AI Visibility Audit does exactly that.
Where to start: diagnose your symptom first
Before working through the layers in sequence, find the one that sounds most like you. That’s where to start.
| If this sounds like you | Start here |
|---|---|
| We rank well on Google but never get named in AI answers | Layers 4 and 5 — trust signals and recommendation eligibility |
| We barely appear at all in ChatGPT, Copilot, or AI Overviews | Layers 1 and 2 — entity foundations and retrieval |
| Our pages are indexed, but generic competitor pages get cited instead | Layer 3 — content extractability |
| We appear in Google, but not in Copilot or ChatGPT Search | Layer 2 — Bing infrastructure specifically |
| We appear but never by name — just as an unnamed source | Layer 5 — recommendation eligibility |
The five layers where AI visibility breaks
Think of AI visibility like a building. Each floor depends on the one below it. You can do excellent work on floor four — trust signals, editorial coverage, third-party citations — but if the ground floor is cracked, none of it holds. The mechanic’s instinct applies here: don’t guess, diagnose. Work through the layers in order.
Layer 1: Entity foundations — can AI systems find and understand you?
Before any AI system can retrieve your content, cite your expertise, or recommend your business, it needs to know you exist as a distinct, identifiable entity in the world — not just as a website. Entity recognition is not the same as being indexed. Google knowing your pages exist is different from AI systems having a confident, consistent model of who your business is.
Here’s a useful analogy. A pub landlord spends six months renovating — new kitchen, new menu, proper real ales. Great local following. Five-star TripAdvisor reviews. But he never updated the Google Maps listing. New opening hours. A phone number that goes nowhere. People ask their phone assistant where to eat nearby. The pub either doesn’t appear or appears with the wrong information. Not because the pub isn’t excellent. Because the information infrastructure behind it is broken. That is the situation most businesses are in with their entity data right now.
Layer 1 checklist
NAP consistency. Your business name, address, and phone number must be identical across your website, Google Business Profile, Bing Places, Apple Business Connect, Companies House, LinkedIn, and every industry-specific directory that matters. Not approximately identical. Identical. A different phone number format on three platforms is enough to undermine entity confidence across all of them.
Wikidata. If you are a notable individual, agency, or organisation, a Wikidata entry with correct properties and sameAs references is one of the highest-value corroboration sources for entity identity. It takes time to do properly and it is worth doing properly. A thin, unsourced entry is worse than none.
Schema markup. Organisation schema on your homepage. Person schema for key individuals. Service schema on relevant service pages. These are the structured declarations that tell AI systems what type of entity you are, what you do, who you serve, and where you operate. If your schema is missing, incomplete, or inaccurate, AI systems will guess. They will guess conservatively.
Apple Business Connect. Overlooked by most businesses. Takes fifteen minutes. Covers Siri, Apple Maps, and Apple intelligence features. For local and professional services businesses, this is a quick win most competitors have skipped.
Companies House and regulated profession listings. For UK businesses, Companies House is a signal AI systems can use to verify your entity exists and is active. For regulated professions — solicitors, accountants, healthcare providers — the relevant regulatory body listing (SRA, ICAEW, CQC) is a strong corroboration signal.
Layer 2: Bing infrastructure — are you visible where B2B buyers are actually searching?
This is the layer most UK B2B businesses are missing entirely. Here is the counterintuitive truth: Google holds roughly 90% of traditional search volume. But ChatGPT Search retrieves from Bing’s index. Microsoft Copilot retrieves from Bing’s index. If you are a B2B business in the UK, a significant proportion of your prospective clients — working in law firms, NHS trusts, local authorities, corporate finance, enterprise IT — are sitting at a Windows machine with Microsoft 365 and Copilot as their default AI assistant. They did not choose Copilot. Their IT department rolled it out. But they are using it. And when they ask it to recommend a supplier, it grounds its response in Bing.
A business perfectly optimised for Google but absent from Bing’s entity infrastructure is invisible at the knowledge retrieval stage for both ChatGPT and Copilot simultaneously. The gap is fixable in an afternoon. Most businesses have never closed it. ChatGPT has 24.9 million monthly UK searches, growing at 83% year-on-year. Microsoft Copilot has 1.5 million, growing at 233%. Bing is the index behind both of them.
Layer 2 checklist
Bing Webmaster Tools. Set it up if you have not. Import from Google Search Console — ten minutes. Check the indexation report: are the pages that matter to you indexed in Bing? Then open the AI Performance dashboard — launched in public preview February 2026 — which shows which of your pages are being selected as grounding sources in Copilot responses. Most businesses have never looked at it. It is the only platform-native diagnostic for Copilot visibility currently available.
Bing Places for Business. The direct equivalent of Google Business Profile for Bing’s ecosystem. When someone asks Copilot to recommend a solicitor, consultant, or service provider, local results draw from Bing Places. If your listing does not exist, has an old address, or shows a changed phone number, you are not on the shortlist.
IndexNow. A protocol supported by Bing that lets you notify search engines instantly when content changes rather than waiting for a crawl cycle. Bing’s crawl cycle is slower than Google’s — IndexNow closes that lag. For WordPress there is a plugin. For other platforms, implementation is straightforward.
Schema validated for Bing. Schema that passes Google’s Rich Results Test can still produce errors in Bing’s pipeline. Run key pages through Bing’s markup validator. Pay attention to Person and Organisation schema, and the sameAs references linking your site to Wikidata and LinkedIn. LinkedIn references are a meaningful entity signal in Bing’s ecosystem because Microsoft owns LinkedIn.
Bing citation directives. The data-snippet attribute steers Copilot toward your best summary paragraphs. NOARCHIVE and NOCACHE directives restrict citation — only use them if you have a specific reason to opt out of Copilot grounding, because doing so removes you from consideration entirely.
Full implementation detail for Bing schema, sameAs specifics, and Copilot grounding is at Why Bing Is Now the Most Important Search Engine for AI Visibility →
Layer 3: Content extractability — can AI systems pull a clean answer from your pages?
AI systems don’t read pages the way humans do. They extract. When an AI system processes your content looking for an answer to a commercial query, it is trying to pull a clean, attributable, standalone response from what you have written. If your pages are structured as long narrative prose with no clear signals — no definitions, no direct answers, no attributable claims — it is not that AI systems cannot read them. It is that they cannot extract from them efficiently, and they move to the page that makes it easier.
Think of it like a reference book versus a novel. You can find information in both. But when you need a specific fact quickly, you reach for the one with headings, definitions, and clear structure. AI systems do the same.
Layer 3 checklist
Opening answer. Every key page should open with a 40–60 word standalone answer to the question that page is designed to answer. Not an introduction. An answer — the kind of paragraph that could be extracted and read in isolation and still make complete sense.
Definitions. Named concepts, frameworks, and technical terms should be explicitly defined, not assumed. “Entity corroboration is the process of…” not “entity corroboration, which we use to…”. The difference matters when AI systems are trying to pull an extractable definition.
Statistics with context and sources. Numbers without attribution carry less weight than numbers with it. “Conversion rates improve with AI citation” is weaker than “Seer Interactive found AI-cited traffic converts at 14.2% versus 2.8% for standard organic — across twelve million visits in 2025.” Source, sample size, date. That is a citable stat.
Attributable claims. AI systems are more likely to cite content that makes specific, named claims they can attribute to a specific author or source. A generic observation gets passed over. A named framework with a stated author and date is attributable, specific, and citable.
Clear H2/H3 structure. Headings should be questions or direct statements, not clever or cryptic. AI systems use heading structure to navigate pages. “Why Bing matters for AI visibility” is more extractable than “The platform you have been ignoring.”
The full page-by-page blueprint for structuring content for AI extraction is at Anatomy of an AI-Citable Page →
Layer 4: Third-party trust signals — do AI systems have independent evidence to trust you?
This is where most businesses fail, and it is the most commercially consequential layer to get right. Your website says you are excellent. Every website says that. AI systems weight it accordingly — as a self-declaration, not as evidence.
What carries weight is what other people say about you — on platforms they control, without incentive to be kind. Editorial, not advertorial. Think about TripAdvisor. Why does a review there carry more trust than a “what our guests say” section on the hotel’s own website? Because you know TripAdvisor has not been paid to say it. The reviewer has no incentive to be kind. That is precisely why it matters. AI systems are learning the same distinction rapidly.
This is the principle that has held true across every era of SEO: you are who you hang with. A link from a trusted, established source transfers credibility in a way a link from your own site never could. Entity corroboration works identically. In a world where AI can generate convincing content at scale — where you genuinely cannot always tell human from machine — trust signals that cannot be manufactured are the only signals that actually count. You cannot fake a decade of genuine editorial coverage. Which is exactly why it is worth building, and why the window for getting ahead of competitors who have not started yet still exists.
Layer 4 checklist
Google reviews. Request them actively and consistently — after genuine positive interactions, not in bulk. Volume and recency both matter.
Clutch (for B2B and professional services). A well-established corroboration source for B2B service provider recommendations. A complete profile with verified reviews is a meaningful trust signal for the kind of high-consideration queries where AI recommendation matters most.
Sector-specific review platforms. Trustpilot for consumer-facing businesses. G2 or Capterra for SaaS. Legal500 or Chambers for law firms. The specific platform matters — sector-appropriate sources carry more weight than generic ones for sector-specific queries.
Editorial coverage. Genuine mentions in industry publications, regional business press, or national outlets — where the journalist chose to write about you, not where you placed a paid feature. This is the hardest signal to build and the most valuable. It cannot be manufactured.
LinkedIn presence. A complete, consistent, active LinkedIn presence is a meaningful entity signal — particularly in Bing’s ecosystem, which weights LinkedIn references directly.
Crunchbase (for B2B and tech businesses). A complete Crunchbase entry is a useful commercial corroboration source, particularly for B2B and tech businesses where AI systems are evaluating vendor credibility.
Layer 5: Recommendation eligibility — the gap between being found and being named
There is a threshold in AI visibility that most businesses never cross. Below it, AI systems know you exist and may reference your content — without naming your brand. Above it, they name you specifically as the recommended choice.
The AI Visibility Ceiling is a diagnostic model developed by SEO Strategy Ltd, not a published metric from Google, OpenAI, or Microsoft — but it reflects a real and commercially important gap between being topically visible and being confidently named. At this layer, the question is simple: is there enough credible evidence, outside your own website, for an AI system to risk naming you?
The commercial consequence of crossing that threshold is significant. Traffic arriving from a named AI recommendation converts at 14.2% according to Seer Interactive’s analysis of twelve million visits. That is not a marginal gain. That is a fundamentally different quality of buyer.
The difference comes down to the depth and consistency of corroboration across layers 1 through 4 working together. No single signal crosses the threshold. The combination — entity data that is consistent everywhere, content that is clearly extractable, third-party trust signals from credible and independent sources — builds the confidence AI systems need to name a business rather than gesture at a category.
The full model — how the seven stages of AI recommendation eligibility work and where most businesses stall — is at How AI Systems Decide Which Companies to Recommend →
What this looks like in practice
| Layer | What you fix there | What it does NOT fix | Time to implement |
|---|---|---|---|
| 1 — Entity foundations | AI systems not recognising your business as a distinct entity | Content quality or trust signals | 1–2 days |
| 2 — Bing infrastructure | Invisible to ChatGPT Search and Copilot despite Google presence | Google AI Overview visibility | Half a day |
| 3 — Content extractability | AI systems unable to pull clean answers from your pages | Third-party trust deficit | Ongoing per page |
| 4 — Third-party trust signals | AI systems not confident enough to name you | Technical entity or retrieval gaps | 3–6 months minimum |
| 5 — Recommendation eligibility | Being mentioned anonymously vs named as the recommended provider | Any single layer in isolation | Compounding over time |
Quick wins by business type
Local businesses. Prioritise NAP consistency across all major platforms, Google Business Profile completeness, Bing Places (most local businesses have never claimed it), Apple Business Connect, and steady review generation on Google and TripAdvisor. These are the signals most weighted for local AI recommendation queries — and most local competitors have only done one or two of them.
B2B companies. Prioritise Bing Webmaster Tools above everything else — open the AI Performance dashboard and understand your Copilot grounding baseline today. Then LinkedIn completeness and activity, Clutch profile with verified reviews, Crunchbase entry, and key commercial page extractability. Your buyers are on Windows estates using Copilot. Bing is the front door and most B2B businesses have never opened it.
Professional services firms. Prioritise regulator listings (SRA, ICAEW, CQC, FCA — whatever applies), person schema for named experts and senior practitioners, authoritative professional bios with explicit specialisms declared, client reviews on sector-appropriate platforms (Legal500, Chambers, Trustpilot), editorial mentions in industry publications, and consistent third-party corroboration of specialism. For regulated professions, the regulator listing is the single highest-weight corroboration signal available — and it is free.
Key definitions
AI Visibility Ceiling: The observable threshold between topical visibility — where AI systems reference your content without naming your brand — and provider visibility, where your business is named as the recommended choice. A diagnostic model developed by SEO Strategy Ltd, not a published platform metric, but reflective of a real and measurable gap between being considered and being recommended. Sean Mullins, SEO Strategy Ltd, March 2026.
Entity corroboration: The accumulation of consistent, independent, third-party evidence about a business entity that increases AI systems’ confidence in naming it as a recommended provider. Distinguished from topical authority — a business can be topically authoritative without being sufficiently corroborated for named recommendation. Sean Mullins, SEO Strategy Ltd, March 2026.
Bing entity infrastructure: The set of Bing-specific signals — Bing Webmaster Tools verification, Bing Places listing, schema validated against Bing’s requirements, sameAs references to Wikidata and LinkedIn — that establish entity identity for retrieval by ChatGPT Search and Microsoft Copilot.
