Being found by AI is important.
Being trusted is essential.
A practical guide to AI search readiness, governance and public trust for UK public-sector organisations.
AI-powered search is changing how people find information about public services. Residents are increasingly turning to tools such as ChatGPT, Google AI Overviews and Microsoft Copilot for answers, rather than navigating directly to council or NHS websites.
For public-sector organisations, this creates a new challenge. It’s no longer enough for official information to appear in search results. It needs to be discoverable, accurate, up to date and trustworthy when AI systems use it to answer people’s questions.
The key takeaway: Being cited by AI doesn’t mean the answer is correct. Public-sector organisations need to measure accuracy and trust, not just visibility.
Why AI search accuracy matters for public trust
Imagine someone asking an AI assistant when their VAT payment is due.
The assistant gives a confident answer. The person follows it without visiting GOV.UK.
But the answer is wrong.
This is not just a hypothetical risk. HM Revenue & Customs identified AI-generated misinformation about VAT payment deadlines falling on weekends and responded with verified alerts to help correct it.
The same challenge exists across public services.
| Sector | Example question | Potential consequence of an incorrect answer |
|---|---|---|
| Local government | Am I eligible for a council tax discount? | Someone misses support or receives incorrect advice |
| Central government | When is my tax payment due? | Missed deadlines or financial penalties |
| NHS and healthcare | How do I access urgent care? | Delayed treatment or inappropriate service use |
| Education | Can I apply for financial support? | Missed eligibility or application deadlines |
| Regulators | How do I renew my registration? | Lapsed registration or inability to practise |
These are not just search queries. They are questions that influence real decisions.
Visibility gets you found. Accuracy builds confidence. Governance protects trust.
What does the new government guidance tell us?
In October 2026, Government Communications published Strengthening public trust in the age of AI-powered search.
The guidance highlights the importance of ensuring that authoritative public information remains discoverable, accurate and trusted as AI changes how people search.
It introduces the PRIME framework:
- Plan: Review your information, identify priorities and address outdated or duplicated content.
- Respond: Write content that answers the questions people actually ask.
- Implement: Apply sound technical practices and clear, structured content.
- Monitor: Understand how AI tools find and represent your information.
- Evaluate: Measure outcomes and continuously improve.
Although the guidance is aimed at government communicators, its underlying principles are relevant to local authorities, NHS organisations, education providers, regulators and other public bodies.
Being cited does not mean being represented accurately
The Office for National Statistics (ONS) provides a particularly important example.
In its assessment of Google AI Overviews, ONS found that:
- 97% of the answers it assessed cited ONS.
- 62% named ONS in the answer text.
- Only 21% were judged fully accurate by its subject matter experts.
These findings relate to the answers and sample assessed by ONS, rather than all AI search results.
The lesson is clear: being cited is not the same as being correct.
A public-sector organisation should therefore measure more than how frequently it appears in AI-generated answers. It should also assess whether those answers accurately represent its information.
What can public-sector organisations actually control?
You cannot dictate how independent AI platforms generate answers. Their systems, sources and responses may change without notice.
However, you can govern the information and processes within your organisation.
That includes:
- Content: Whether information is accurate, current, accessible, clear and authoritative.
- Technical discoverability: Whether search engines and permitted AI crawlers can access relevant pages.
- Ownership: Who is responsible for maintaining important information.
- Monitoring: How inaccurate or misleading AI-generated answers are identified.
- Response: How errors are investigated, corrected, escalated and reviewed.
This is where search engine optimisation (SEO), generative engine optimisation (GEO), answer engine optimisation (AEO) and AI governance meet.
SEO supports discoverability in conventional search. GEO and AEO focus on making information easier for AI-powered systems to find, interpret and reference. AI governance provides the accountability, risk management and oversight needed to make those efforts responsible.
None of these disciplines guarantees that an external AI platform will produce a correct answer. Together, however, they give organisations a stronger foundation for protecting public trust.
Choose five high-impact service questions, compare AI answers with your official guidance, and record what you find. Go straight to the testing method.
10 practical AI search checks you can carry out today
You do not need to buy a specialist AI monitoring platform to get started.
The following checks use widely available tools, existing website data and straightforward testing methods.
Short on time? Start with Check 6. Testing five important public-service questions can give you a useful first indication of your risks.
See whether AI tools cite your website using Microsoft Clarity
Tool: Microsoft Clarity
Microsoft Clarity offers AI Visibility features that can help website teams understand how their content is referenced in supported AI-generated answers.
What to do
- Check whether your organisation already has a Microsoft Clarity project.
- If it does, open AI Visibility → Citation.
- If access or verification is required, follow the available domain-verification instructions.
- Review which pages are cited and the topics or grounding queries associated with them.
- Identify important service pages that deserve an accuracy review.
Grounding queries are searches an AI system uses when retrieving information to support an answer. They are not necessarily the exact words a person originally typed.
What to look for
Which pages receive citations? Are they current? Do they contain authoritative information? Are important services missing from the available data?
What to do next
Add frequently cited pages and relevant queries to your audit register. Prioritise accuracy, particularly where an incorrect answer could cause harm.
Important: Clarity’s citation data covers supported AI experiences and should not be treated as a complete record of every AI-generated answer.
Clarity is separate from Microsoft 365. If you do not already use it, assess privacy, consent and tracking requirements before introducing it.
Identify visitors arriving from AI assistants in Google Analytics 4
Tool: Google Analytics
Google Analytics 4 (GA4) can help you understand which visitors arrive through identifiable links from AI assistants.
What to do
- Open your GA4 property.
- Review Reports → Acquisition → Traffic acquisition.
- Examine the available session source and referral information.
- Look for identifiable AI referrers, such as ChatGPT, Perplexity or Copilot.
- Where appropriate, create a custom channel group for AI assistant traffic.
Ask your analytics lead to check Google’s current channel-group guidance before configuring rules, as available settings and referral patterns can change.
What to look for
Which service pages receive AI-referred visits? Are visitors reaching the right information? Do they continue to application forms or other relevant actions?
What to do next
Review the landing pages and compare the information they contain with the answers AI assistants give.
Remember: GA4 measures identifiable visits, not every AI citation or answer. People may receive an answer without clicking through.
For public services, a correct answer that helps someone complete a task without visiting a website can be a positive outcome.
Review your robots.txt file
Tool: Your web browser
A robots.txt file provides instructions to web crawlers about which parts of a website they may access.
Some organisations restrict AI crawlers without fully understanding the difference between search discovery and model training.
What to do
- Open your organisation’s website.
- Add
/robots.txtto the end of the domain. - Look for rules that apply to search engines and AI crawlers.
- Ask your technical team to confirm whether any restrictions are intentional.
For example, OpenAI uses different crawler identities for different purposes:
| Crawler | Purpose |
|---|---|
| OAI-SearchBot | Supports discovery for ChatGPT search |
| GPTBot | Used for model training-related crawling |
An organisation may decide to permit search crawling while restricting training-related crawling.
For illustration, the following rules express that distinction:
User-agent: OAI-SearchBot
Allow: /
User-agent: GPTBot
Disallow: /
Do not paste these rules into your live website without technical review. They are illustrative, not a complete robots.txt configuration.
What to look for
Check whether important pages are unintentionally blocked. Also consider firewall and content delivery network (CDN) rules, which may restrict crawlers separately.
What to do next
Document your crawler policy with technical, security and information governance colleagues.
Allowing crawling does not guarantee indexing, citations or accurate AI answers.
Check whether Google can index your important pages
Tool: Google Search Console
What to do
- Open Search Console for your organisation’s website.
- Select five important service pages.
- Paste each URL into the URL Inspection tool.
- Check whether the page is indexed and whether Google reports problems.
- Review relevant search performance reports, including AI search reporting where available.
What to look for
Potential issues include:
- Pages excluded by a
noindexdirective. - Pages blocked from crawling.
- Duplicate pages with conflicting canonical URLs.
- Important pages that are not indexed.
What to do next
Ask your website team to investigate the cause and correct any unintended restrictions.
Search Console can help assess Google’s visibility, but it does not tell you whether every AI-generated answer is accurate.
Review Bing discoverability and AI citations
Tool: Bing Webmaster Tools
Bing Webmaster Tools provides indexing and search diagnostics. Its AI-related reporting can also provide citation insights where supported.
What to do
- Open or create your verified Bing Webmaster Tools property.
- Inspect priority service URLs.
- Confirm your sitemap has been submitted.
- Review indexing issues.
- Check whether AI Performance reporting is available for your property.
What to look for
Are your important pages discoverable? Which pages receive AI citations in the supported experiences? Are older pages appearing instead of current guidance?
What to do next
Resolve indexing problems and include relevant findings in your AI search audit.
Again, citations are useful indicators of visibility, not proof of factual accuracy.
Test 20 real questions people ask about your services
Tools: ChatGPT search, Google AI search, Microsoft Copilot and other relevant AI search services.
This is the single most important activity in this guide.
Dashboards tell you something about discoverability. Testing real questions helps you understand what people may actually be told.
What to do
Start by identifying 20 questions from:
- Your website’s internal search terms.
- Contact centre enquiries.
- Complaints and service feedback.
- User research.
- Search Console queries.
- Common questions received by frontline teams.
Prioritise questions involving eligibility, payments, deadlines, health, safety and important service decisions.
For each question:
- Find the current authoritative answer on your website.
- Ask the question in three AI search tools.
- Record the date, question, answer and cited sources.
- Check the answer against the official source.
- Repeat the test to identify inconsistencies.
- Ask a relevant subject matter expert to review potentially harmful inaccuracies.
Use a consistent testing method and record whether tools are signed in, have web search enabled or use location information. AI answers can vary with context and time.
How to rate the results
| Rating | Meaning |
|---|---|
| Accurate | All important facts are correct |
| Partly accurate | Broadly correct, but an important detail is missing |
| Inaccurate | Material information is incorrect, unsupported or outdated |
| No usable answer | The tool does not provide enough information to act |
Pay particular attention to material facts: eligibility, costs, deadlines, exceptions and the correct next step.
A response that cites your website but gives the wrong eligibility criteria should not be considered a success.
Validate structured data
Tools: Google Rich Results Test and Schema Markup Validator
Structured data is machine-readable information that helps describe a webpage’s content.
For public-sector websites, relevant schema types may describe an organisation, service, location or opening hours.
What to do
- Choose a priority page.
- Paste its URL into the validation tools.
- Review detected structured data and reported errors.
- Compare the markup with the information visible to users.
What to look for
Incorrect organisation details, outdated opening hours, missing required fields for supported features or markup that contradicts the page.
What to do next
Ask your development team to correct relevant errors and ensure structured data accurately represents the content.
Structured data is useful, but it is not a guaranteed route into AI-generated answers.
Find outdated, duplicated or contradictory information
Tools: Your content management system, website search and Google.
AI systems may encounter old PDFs, archived pages or conflicting information alongside your current guidance.
What to do
Search for a priority topic across your website.
For example:
site:example.gov.uk council tax discount
Review the results and identify every page that provides relevant guidance.
What to look for
- Multiple pages giving different answers.
- Outdated fees, eligibility criteria or deadlines.
- Old PDFs that remain discoverable.
- Pages without clear ownership.
- Important answers buried beneath unnecessary introductory text.
What to do next
Agree which page should be the authoritative source. Correct or consolidate conflicting content, and redirect or remove outdated material where appropriate.
Use GOV.UK content design guidance to make information clear, accessible and task-focused.
A useful content rule: Answer the question first, then explain the details, exceptions and next steps.
Understand which AI crawlers access your website
Tools: Hosting logs, CDN analytics or supported Microsoft Clarity Bot Activity features.
What to do
Ask your technical or hosting team whether they can identify requests from known AI crawlers.
Where supported, review crawler names, request volumes and the pages accessed.
What to look for
Are permitted crawlers reaching priority pages? Are there unexpected access restrictions or large volumes of automated requests?
What to do next
Investigate unexplained restrictions with your technical team.
Microsoft Clarity’s Bot Activity feature requires an appropriate infrastructure integration; it is not automatically available simply because Clarity is installed.
Keep these four concepts separate:
Crawling → Indexing → Citation → Website visit
Each describes a different activity. None guarantees the next, and none proves an AI answer is correct.
Create an AI search governance register
Tools: Microsoft Excel, SharePoint Lists or an existing organisational risk-tracking system.
You do not need a complicated governance platform to start.
A simple register makes the work repeatable and ensures someone is accountable.
Recommended fields
| Field | What to record |
|---|---|
| Priority question | The question people ask |
| Service area | The service affected |
| AI platform | Where the answer appeared |
| Date tested | When it was checked |
| Official source | The correct authoritative page |
| Citation status | Whether the official source was cited |
| Accuracy rating | Accurate, partly accurate, inaccurate or no usable answer |
| Potential impact | What could happen if someone follows the advice |
| Risk rating | High, medium or low |
| Corrective action | What needs to change |
| Responsible owner | Who is accountable |
| Retest date | When the issue will be checked again |
| Outcome | Whether the problem improved |
Review high-risk items promptly and agree a regular monitoring cycle.
For each identified problem, distinguish between an error in your own content and an error introduced by the AI platform. The appropriate corrective action may be different.
Worked example: checking an AI answer about council tax
The following example is illustrative. The organisation, AI responses and outcomes are fictional and demonstrate the audit method rather than actual research findings.
A local authority identifies a common question from its contact centre:
“I live alone. Can I get a council tax discount, and how do I apply?”
Its authoritative website explains that a qualifying adult living alone can receive a 25% single-person discount, provides the application route and explains the responsibility to report relevant changes.
The digital team tests the question across four AI tools.
| AI tool | Example response | Assessment |
|---|---|---|
| Tool A | Explains the 25% discount and correct application route, but omits reporting changes | Partly accurate |
| Tool B | Says the discount is 50%, despite citing the council website | Inaccurate |
| Tool C | Directs the person to an unrelated commercial application website | Inaccurate |
| Tool D | Says only to contact the council | No usable answer |
The team discovers that an older council webpage contains different discount categories in a confusing layout.
It takes the following actions:
- Corrects the content: Adds a clear answer about the single-person discount at the top of the authoritative page.
- Clarifies the process: Makes the application link and reporting responsibilities prominent.
- Removes ambiguity: Separates different discount categories under clear headings.
- Addresses outdated material: Reviews old pages and PDFs for removal or redirection.
- Records the risk: Logs the inaccurate AI responses, their sources and potential impact.
- Reports errors: Uses available AI platform feedback mechanisms.
- Retests: Checks the same questions at agreed intervals and records the outcomes.
The team cannot guarantee that external AI platforms will immediately change their answers.
But it now has a clear source of truth, documented evidence, accountable owners and a repeatable process.
The same method works for NHS service information, student finance guidance, professional registration requirements and other public-sector services.
Your first 30 days: a practical AI search trust and readiness plan
You don’t need to audit your entire website at once.
Start with 20 important questions, a small cross-functional team and a manageable review process.
The following plan is designed to establish a baseline and address the highest risks within 30 days. It is not a promise that every external AI answer can be corrected within that period.
Week 1 — Identify your priorities
Your objective: Understand which questions matter most.
Actions
- Nominate a senior accountable owner and a day-to-day lead.
- Gather common questions from website search, contact centres, complaints and user research.
- Select 20 priority questions.
- Rank them by the potential harm caused by incorrect advice.
- Create your governance register.
People involved: Digital, communications, content, service owners and frontline teams.
Deliverable: A prioritised list of 20 questions, with authoritative sources and named owners.
Week 2 — Establish your baseline
Your objective: Understand what AI tools currently tell people.
Actions
- Test each priority question across three relevant AI search tools.
- Record accuracy, citations and material omissions.
- Inspect important pages in Google Search Console and Bing Webmaster Tools.
- Review robots.txt and other relevant crawler restrictions.
- Check existing AI citation or referral reporting where available.
- Ask subject matter experts to review high-risk findings.
People involved: Content, analytics, technical colleagues and service specialists.
Deliverable: An initial accuracy and visibility assessment, with a prioritised list of risks.
Week 3 — Correct the highest-risk issues
Your objective: Improve the authoritative information and remove avoidable barriers.
Actions
- Fix inaccurate or ambiguous source content.
- Put direct answers, eligibility and next steps prominently on relevant pages.
- Review conflicting pages and outdated PDFs.
- Address unintended indexing or crawler restrictions.
- Validate relevant structured data.
- Assign content owners and review dates.
People involved: Content designers, service owners, developers and information governance colleagues.
Deliverable: Published improvements and an action plan for unresolved issues.
Week 4 — Retest and establish governance
Your objective: Make AI search accuracy an ongoing responsibility.
Actions
- Repeat the original questions using the same testing method.
- Compare findings with the baseline.
- Record which inaccuracies remain unresolved.
- Report serious errors to AI platforms where possible.
- Agree how potentially harmful misinformation will be escalated.
- Establish a regular monitoring cycle.
- Share a short scorecard with the accountable owner.
People involved: Digital, communications, service owners and relevant governance leads.
Deliverable: A simple AI search trust scorecard and an agreed monitoring process.
Remember that AI platforms may take time to reflect content changes. Record both improvements and unresolved issues without assuming that a particular content update caused an observed change.
Your AI Search Trust and Readiness Checklist
Use this checklist as a starting point for your organisation.
Content and authority
- We have identified our 20 highest-priority public-service questions.
- Each question has an accurate, authoritative source.
- Important pages answer questions clearly in plain English.
- Outdated, duplicated and contradictory content is being addressed.
- Priority content has an owner and review date.
Technical discoverability
- Important pages have been checked in Google Search Console.
- Important pages have been checked in Bing Webmaster Tools.
- Our robots.txt rules have been reviewed.
- Relevant firewall and CDN restrictions have been considered.
- Structured data has been checked where appropriate.
Accuracy and public trust
- We have tested priority questions across multiple AI tools.
- We record whether cited sources support the answers given.
- We distinguish accurate, partly accurate and inaccurate answers.
- Subject matter experts review potentially harmful inaccuracies.
- We have a process for reporting and escalating serious errors.
Governance and monitoring
- A senior accountable owner has been nominated.
- We maintain an audit and corrective-action register.
- We prioritise issues according to potential user harm.
- We have agreed a regular retesting process.
- Findings are reported to the appropriate governance leads.
How should you measure success?
Avoid measuring AI search readiness solely through rankings, citations or website visits.
For public services, the most important outcome is whether people receive trustworthy information.
Start with five measures.
| Measure | What it tells you |
|---|---|
| Answer accuracy rate | How many tested answers contain all material facts correctly |
| High-risk inaccuracies | How many potentially harmful problems remain unresolved |
| Authoritative citation rate | How often tested answers cite the appropriate official source |
| Ownership coverage | How many priority pages have named owners and review dates |
| Corrective-action progress | How effectively identified problems are investigated and addressed |
Record the number of questions, platforms and test runs behind each measure so that comparisons remain meaningful.
You can also monitor AI referrals and citation trends, but these should support—not replace—accuracy and risk measures.
How does this support AI governance and UK public-sector standards?
AI search readiness is not a substitute for a wider AI governance framework.
However, the activities described in this guide support several recognised principles of responsible AI.
The UK Government’s five AI regulatory principles, developed through the Department for Science, Innovation and Technology (DSIT), provide a useful reference.
| Principle | Practical application |
|---|---|
| Safety, security and robustness | Prioritise potentially harmful misinformation |
| Appropriate transparency and explainability | Review sources, citations and the evidence behind answers |
| Fairness | Test different ways people ask questions and consider accessibility |
| Accountability and governance | Assign owners, document decisions and monitor risks |
| Contestability and redress | Establish ways to identify, report and address errors |
These are regulatory principles, not a single standalone AI law imposing identical obligations on every organisation.
Other relevant frameworks include:
- Government Data and AI Ethics Framework: Supports responsible data and AI practices.
- GDS Service Standard: Provides a framework for designing and operating user-centred services, with application depending on the service and organisation.
- Public-sector accessibility requirements: Help ensure that digital information and services are accessible.
- UK GDPR and the Data Protection Act 2018, as amended: Relevant where personal data is processed, including through analytics or AI systems.
- Algorithmic Transparency Recording Standard: Establishes transparency expectations for relevant public-sector algorithmic tools, with mandatory requirements applying to specified central government organisations and in-scope uses.
- ISO/IEC 42001: An international standard for AI management systems, supporting defined responsibilities, risk assessment, oversight and continual improvement.
Not all of these requirements apply to every organisation or every activity. In particular, independently operated AI search platforms are different from AI systems a public body deploys or controls itself.
The important principle is to apply governance proportionately, based on organisational responsibilities and the risks involved.
For externally generated AI answers, this means maintaining reliable source information, monitoring risks and responding appropriately.
For AI services your organisation operates itself, governance must also address the system’s design, data, testing, oversight and operation.
Where should your organisation start?
If you only do one thing after reading this guide, do this:
Choose five important questions about your services. Ask them in three AI search tools. Compare the answers with your official information.
Record what you find.
You may discover that your content is being cited accurately. You may find that a critical detail is missing. Or you may uncover an outdated page that needs attention.
Whatever the outcome, you will have taken the first practical step towards understanding and governing your organisation’s AI search presence.
The objective isn’t simply to appear more often in AI-generated answers.
It is to help people receive accurate, accessible and trustworthy information about the public services they rely on.
How Invuse can help
We work with UK public-sector organisations to improve digital services, strengthen content quality and support responsible AI adoption.
Our expertise spans AI Search Readiness (GEO/AEO), Content and Conversational Design, User Research, Accessibility, Digital Optimisation and AI Governance and Assurance.
We also hold ISO/IEC 42001 certification for our AI management system, reflecting our commitment to structured, accountable AI governance.
Whether you’re beginning to investigate how AI platforms represent your organisation or looking to establish a more comprehensive governance approach, we can help you turn the principles into practical action.
This guide references Government Communications guidance, UK public-sector frameworks and publicly available technology documentation. Product capabilities and regulatory requirements can change; organisations should check current guidance and assess their own circumstances.
Find out how AI search represents your organisation
AI-powered search is already shaping how people discover public services. But is your organisation's information being found, cited and represented accurately? Request an initial AI search readiness review from Invuse. We'll look at how your organisation appears in AI-generated answers and identify potential opportunities to improve visibility, accuracy and trust.Stay informed
Practical guidance on responsible AI, accessibility and digital transformation for the public sector, delivered to your inbox.