AI Visibility Audit Deliverables: A UK Buyer's Specification

A credible paid AI visibility audit should answer seven buying questions:

1. Can relevant search and answer systems access the right evidence?
2. Do they understand which company, people, products and locations the evidence describes?
3. Does the brand appear accurately in a controlled sample of buyer questions?
4. Which owned and external sources are cited?
5. Which competitors and source types currently shape the answers?
6. What prevents the business from being a useful, corroborated option?
7. Which fixes should happen in the next 30, 60 and 90 days?

If the audit provides only a score and a list of prompts, it has measured observation without readiness, cause or action.

This is a procurement specification: use it to compare a provider's scope, evidence and handover. If you want to run the checks yourself, use the separate DIY AI citation audit guide. Keeping those intents separate prevents a supplier checklist from becoming another generic prompt tutorial.

Checked: 14 July 2026 · Use this specification when buying or reviewing an AI search audit

The required audit sections

You should receive the raw material behind every summary chart.

Use the audit acceptance worksheet

Download the AI search audit acceptance worksheet. It turns the sections above into a review record: required evidence, observed result, evidence location, owner, acceptance decision and remaining limitation. The result fields are blank because this is a template, not a completed audit or a visibility score.

Ask the provider to complete it against the agreed scope. For example, an answer screenshot should identify the exact question, platform, mode and date, while a recommended page change should have an owner and a check that shows whether it was implemented. Keep private customer information and credentials out of shareable exports.

When comparing Search Console with analytics, record the period, time zone, country, channel and whether the data is provisional. Search impressions, clicks, sessions and confirmed enquiries are different measurements. A last-24-hours snapshot is useful for spotting a question to investigate; it is not, on its own, evidence of a sustained uplift or a project's return.

1. Audit scope and methodology

The first page should state:

• organisation and brands reviewed;
• products, services and locations;
• UK or other target markets and languages;
• buyer stages and prompt groups;
• AI products, modes and versions where identifiable;
• logged-in, location and personalisation conditions where relevant;
• run dates and repetition method;
• competitor selection;
• known reporting limitations.

Answers change. The audit cannot freeze the systems, but it can make the sample auditable.

What makes a useful prompt set?

The set should cover real buying questions:

• category discovery;
• specific problems;
• cost and suitability;
• alternatives and comparisons;
• local or regional selection;
• implementation risk;
• brand facts and proof.

Inputs should include Search Console, sales questions, customer research, site search and competitor language. A generated list of minor wording variations is not market research.

For a small or focused business, 25–50 carefully selected prompts can be enough for a baseline. Larger portfolios may need more, but the reviewer must still inspect answer quality and citations rather than only count mentions.

2. Technical eligibility

The audit should test the live website, not rely on a CMS screenshot.

Google eligibility

Google says supporting links in AI Overviews and AI Mode must come from pages that are indexed and eligible to show a search snippet. Its AI feature documentation also says there are no special technical requirements, AI schema or machine-readable AI files required.

Check:

• status codes and indexability;
• canonical URLs;
• Google-selected canonical for priority pages where available;
• robots and CDN access;
• important content in rendered text;
• internal links and sitemap discovery;
• structured data matching visible information;
• duplicate or conflicting pages;
• Business Profile and Merchant Center accuracy where applicable.

ChatGPT search eligibility

OpenAI's publisher FAQ recommends allowing OAI-SearchBot when a publisher wants content included in ChatGPT summaries and snippets.

Check the production robots.txt, CDN, firewall and response behaviour for the crawler. Do not stop at whether a line appears in a text file.

Other systems

Document which crawlers and policies were checked and why. Avoid enabling every bot by reflex. Access should follow the business's discovery, training, licensing and privacy choices.

3. Entity and factual accuracy

Create a canonical fact sheet:

• legal and trading names;
• brand relationships;
• founders, authors and experts;
• services and products;
• locations and service areas;
• contact and website details;
• price or availability facts the business chooses to publish;
• accreditations, awards and memberships;
• verified proof and case studies.

Compare it with the website, structured data, profiles, directories and sources appearing in AI answers.

Classify discrepancies:

A mention attached to the wrong business is negative visibility.

4. Sampled answer visibility

For each prompt, record:

• exact wording and follow-up;
• engine, mode, date and relevant settings;
• brands appearing;
• the role each brand plays;
• cited URLs;
• factual accuracy;
• whether the answer addresses the buyer's constraint;
• material omissions or unsafe claims.

Report by prompt group. A company may be strong for brand facts and absent from discovery and comparison questions. One overall percentage hides that.

Useful classifications:

• recommended option;
• compared option;
• source cited;
• passing mention;
• inaccurate mention;
• mistaken identity;
• not present;
• prompt not eligible or answer unavailable.

The audit should preserve full observations or enough evidence for the client to verify the classification within copyright and platform rules.

5. Citation and source analysis

List every recurring cited domain and URL with:

• owner: brand, independent, competitor or platform;
• source type: documentation, article, directory, forum, review, profile, dataset or media;
• prompt groups where it appears;
• freshness and factual accuracy;
• whether the business controls, can influence or can only monitor it;
• the next reasonable action.

The audit should answer:

• Which owned pages are already cited?
• Which third-party sources define the category?
• Are old pages or facts still influential?
• Do competitors earn citations through better documentation, stronger proof or wider corroboration?
• Which source gaps can be addressed ethically?

The answer is not “build backlinks to everything”. A source may be influential because it contains original data, clear comparison or community experience.

6. Content and proof gaps

Map buyer questions to the current page that owns them.

Decisions should be one of:

• improve existing page;
• consolidate overlapping pages;
• create a distinct new page;
• add primary evidence;
• correct an external profile;
• produce a source-worthy asset;
• no action because the opportunity is weak.

High-value gaps commonly include:

• price and cost drivers;
• suitability and exclusions;
• comparisons and alternatives;
• process and implementation risk;
• original results or methodology;
• location and availability;
• named expertise and review dates;
• references to authoritative primary sources;
• clear next steps for a qualified buyer.

The audit should check cannibalisation before recommending content. Five pages defining AEO do not create five times the authority.

7. External corroboration and authority

Assess whether important claims are supported outside the site:

• official industry or professional profiles;
• partner documentation;
• credible customer proof;
• relevant directories with editorial standards;
• expert contributions;
• independent research and datasets;
• publications and communities used in sampled citations.

Separate correction, contribution, earned coverage and paid placement. The audit should never disguise bought mentions as independent validation.

An authority plan should begin with something worth referencing: original data, an excellent implementation guide, a transparent comparison, a useful tool or genuine subject expertise.

8. Measurement baseline

The audit needs a starting point for future work.

Controlled sample

Report accurate brand inclusion, citation inclusion, source mix and errors by prompt group and engine. Keep raw prompts and dates.

Google Search Console

Record relevant page and query performance. Check access to Google's Generative AI performance report, which has a limited rollout. If it is unavailable, retain ordinary Web reporting and explain that those totals do not isolate AI activity.

Analytics referrals

Record identifiable ChatGPT, Perplexity and other AI referrals, landing pages and actions. Explain consent, dark-referral and sample-size limits.

Commercial baseline

Record qualified enquiries, calls, sales and reported discovery sources where the business can do so reliably. Add a simple “first heard / checked before contacting” question to the CRM or enquiry process.

The full measurement method is in how to measure AI search visibility. The audit should establish the baseline, not pretend one month proves return.

9. Prioritised action plan

Each recommendation needs:

Order the plan:

First 30 days

Access, indexability, mistaken identity, material factual errors and measurement gaps.

Days 31–60

Priority commercial pages, entity clarity, internal linking, structured data and owned profiles.

Days 61–90

Original proof, credible source development, controlled resampling and roadmap updates.

An action plan without owners is a wish list.

What should not pass as an audit?

• a score with no raw prompts;
• screenshots with no engine, date or exact question;
• only branded prompts;
• a robots and llms.txt check presented as the technical audit;
• schema recommendations without inspecting visible content;
• a content-volume plan without page ownership or cannibalisation review;
• a backlink package presented as authority;
• no factual accuracy assessment;
• no Search Console or commercial baseline;
• guaranteed citations;
• a 90-page PDF with no ordered implementation list.

What should an AI visibility audit cost?

Ask for a scoped quote that identifies the markets, engines, prompts, pages and technical checks included. Separate the assessment from implementation and ongoing monitoring. A price alone does not tell you whether the evidence or the resulting action plan will be useful.

Our AEO pricing UK guide separates audits from monthly implementation and explains what higher budgets should add.

Questions to ask before buying

1. Which exact engines, modes and markets are included?
2. How many curated prompts and which buyer stages?
3. Will we receive the raw answers and cited URLs?
4. How do you assess accuracy and mistaken identity?
5. Which technical checks run against the live site?
6. How do you avoid duplicating our SEO audit?
7. Will recommendations include owners and acceptance tests?
8. What data can current platforms not expose?
9. What can we export and retain?
10. Is implementation included or separately priced?

Frequently asked questions

What is an AI visibility audit?

It is a structured review of whether answer systems can access and understand a business, how the brand appears across a controlled prompt sample, which sources are cited and what technical, content or authority work should happen next.

Is an AI visibility score reliable?

Only as a summary of a disclosed sample. Ask for prompts, engines, dates, weighting, raw answers and accuracy classifications. It is not a complete measure of every answer a buyer may see.

Does the audit need access to Search Console?

Yes for a useful search baseline and index evidence, though a preliminary external check can begin without it. Analytics and CRM access improve the commercial baseline.

Should an audit recommend llms.txt?

It may discuss the file in context, but Google says no special AI text file is required for its AI features. A credible audit prioritises indexable evidence, clarity and corroboration.

How often should the audit be repeated?

Do the deep baseline once, implement it, then monitor a controlled set monthly and review the wider audit quarterly or after a major site, brand or market change.

Buy the diagnosis only if it can become work

The audit should leave you with evidence you can inspect, priorities you can assign and checks that prove completion. Anything less is a visibility demo.

Related: AI search optimisation service · Measure AI search visibility · Choose an AI search agency · AEO pricing UK