# Clients Are Asking About AI Visibility: What Goes in the Agency Report

> Learn how to report AI search visibility with buyer prompts, competitor context, source evidence, and clear next actions.

Source: https://www.citedintel.com/ai-seo/united-kingdom/ai-visibility-client-reporting-what-agencies-put-in-the-report
Published: 2026-09-18
By Cited (citedintel.com) — the generative engine optimization (GEO) platform.

---

A monthly AI search report should not copy a conventional ranking report. It should show whether a client is mentioned, recommended, supported by useful sources, and visible against the companies buyers actually compare.

AI search optimization reporting turns scattered assistant answers into a decision record for the agency and the client. A useful section connects observed answers with first-party exposure, recommendation context, source quality, and the next work that can change the result.

## The agency report should answer a buying question

Clients asking about [AI search visibility](https://www.citedintel.com/answer-engine-optimization/own-data-research-2026-w37) usually want a yes-or-no answer: are buyers finding us when they ask an AI search engine for a supplier? A credible agency report gives a more useful answer by separating presence, suitability, evidence, and competitive share.

The report should separate evidence types instead of compressing them into one score.

My view is simple: report evidence, movement, and decisions. A page of screenshots can show that an assistant named a company, but it cannot show whether the company was suitable for the buyer’s requirements or whether the answer relied on current information.

## Build the UK view from two evidence layers

A UK client report should separate Google’s first-party exposure data from observations in third-party AI assistants. Google Search Console’s generative AI reporting includes impressions, appearing pages, countries, devices, and performance over time, and Google said the report reached every website worldwide on 31 August 2026 in its [Search Central update](https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports).

That distinction gives the account lead a useful reporting structure. Search Console can show whether the client’s pages were exposed in Google’s generative features for people in the United Kingdom. Prompt audits can show how answer engines describe, compare, or support the client when a buyer requests a supplier.

- **Owned exposure:** pages, countries, devices, impressions, and performance recorded in Google’s generative AI report.
- **Observed presence:** whether an assistant names the client within a defined buyer-question set.
- **Recommendation context:** whether the assistant presents the client as suitable for the stated requirements.
- **Source support:** which pages, reviews, analyst references, or other public sources support the answer.

For a UK legal technology client, the prompt set should use the language its buyers use locally. “Best legal case management software for a UK firm” gives the agency more useful evidence than “best legal software” because the answer may need to account for UK terminology, regulatory expectations, data hosting, court workflows, and integrations.

Country belongs in the reporting table, not only in the campaign brief. A company can lead on a US question and disappear from a UK question because local terminology, providers, public evidence, or cited sources differ.

Google’s guidance, updated in July 2026, says AI search features rely on core Search systems and related searches, while meeting best practices does not guarantee crawling, indexing, or inclusion. Place that caveat beside the methodology so the client understands why a technical SEO audit and an answer audit measure separate conditions.

## Do not combine mentions, recommendations, and citations

A mention records that a company was named in an answer. A recommendation records that the assistant positioned the company as suitable for a buyer’s needs. A citation records the public source used to support a claim, so the three outcomes need separate rows and definitions.

| Signal | Meaning for the client | Evidence to display | Useful agency reading |
| --- | --- | --- | --- |
| Mention rate | How often the client was named within audited answers | Prompt, engine, date, and answer excerpt | Basic presence exists, but selection may still be weak |
| Recommendation rate | How often the client was presented as suitable | Buyer requirement, alternatives named, and answer position | The client entered a genuine buying conversation |
| Citation rate | How often public evidence supported the answer | Cited URL, source type, and claim supported | Source quality and factual context need review |
| Share of voice | How often the client appeared relative to relevant rivals | Rival set, prompt class, and market | Competitive standing becomes visible |

Share of voice only helps when the rival set reflects the client’s sales reality. For UK legal tech, the set may include established practice-management vendors, specialist case-management products, and an existing system or internal-build option. The account team should ask sales colleagues which alternatives turn up in lost deals, then use those names in the audit.

Do not label a company “number one” because an answer placed it first once. OpenAI’s September 2026 Search guidance says ChatGPT may rewrite a question into targeted searches and use search partners to retrieve information. Repeated audits provide a stronger basis for reporting movement than a single answer position.

A useful chart can show recommendation share across four prompt groups: category discovery, compliance requirements, switching, and competitor comparison. Keep each group’s denominator visible. A legal intake product may perform well for “software for legal client intake” but poorly for “UK legal intake software with secure document collection”. Those findings point to different work.

## The monthly section needs a fixed reporting frame

A credible monthly AI search section contains a defined audit record, meaningful metrics, answer excerpts, source evidence, and one named next action. The section should help an account lead decide what to publish, repair, investigate, or stop doing.

### Record the conditions before showing the result

Put the reporting frame at the top of the page. State the month, market, language, location setting, engines checked, prompt count, prompt groups, and comparison set before showing the scorecard.

- **Period:** 1 to 30 September 2026, or the agency’s stated audit window.
- **Market:** United Kingdom, with England, Scotland, Wales, and Northern Ireland distinctions when the service differs.
- **Language:** English, plus Welsh or another language when the client serves those searches.
- **Location:** UK location setting used for each audit.
- **Engine:** record each source separately rather than blending every result into one AI channel.
- **Prompt groups:** category, requirements, comparison, switching, pricing, and implementation.
- **Rivals:** names taken from current sales conversations and actual UK alternatives.

The engine field should name ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews as separate rows in the methodology, not as one blended “AI” channel. Citedintel audits buyer-intent prompts across those environments and can add Google AI Overviews or another engine for Enterprise work when the reporting requirement calls for it.

### Show movement with enough context to judge it

The scorecard should show this month, the previous audit, and the change in percentage points. Include the number of prompts behind every figure, because a movement based on six questions deserves less weight than movement across a stable, larger set.

- **Mentions:** client mentions divided by audited answers.
- **Recommendations:** client recommendations divided by answers where a supplier or product was requested.
- **Citations:** answers with at least one supporting client source, separated from uncited mentions.
- **Share of voice:** client recommendation share against the agreed rival set.
- **Position:** answer location, such as first suggestion, later suggestion, or comparison-only presence.
- **Movement:** change since the previous audit, with both audit dates visible.

Do not compress these signals into an unexplained “AI visibility score”. A single number can hide a valuable citation improvement behind a fall in mentions, or make a weak recommendation look healthy because the company name occurred in a long list.

### Put the answer beside the number

Every important metric needs an evidence panel containing the prompt, answer excerpt, date, engine, location, and cited URLs. A client should be able to inspect the result without taking a colored chart on trust.

For a legal tech account, show the sentence that names the vendor and the sentence that explains the fit. If an answer recommends the client for small firms but says nothing about UK data handling, preserve that limitation in the report. The missing qualification may matter more than the name itself.

Source quality needs its own label. A recommendation supported by a current product page is useful, but a recommendation supported by an outdated directory listing, an incorrect review, or an unrelated page creates a correction task.

## An absent client gives the retainer a specific job

An absent client is not a failed report. It is a documented gap between a buyer question and the public evidence available to answer it, which gives the agency a specific retainer conversation.

Show absence by prompt class rather than as one alarming total. “Not recommended for UK firms needing legal document automation” is more useful than “low AI search visibility” because the first finding points to a commercial requirement and a content decision.

Use a gap card with five fields:

- **Buyer question:** the precise UK prompt that produced no client mention.
- **Rival answer:** the vendors or alternatives that did appear.
- **Missing signal:** the requirement a rival addressed and the client did not support publicly.
- **Evidence location:** the client page, review profile, document, or third-party source that should answer the requirement.
- **Next action:** one owner, one asset, and one recheck date.

Suppose a UK legal operations buyer asks for case-management software with matter-level reporting and Microsoft integration. The report should show whether recommended rivals have a page, review, or independent reference stating those details. The agency can then propose a product comparison section, a technical integration page, or a third-party evidence project instead of selling “more content” as an abstract remedy.

I keep telling account leads to put the gap before the recommendation. A client can debate a proposed article. It is harder to dismiss a dated answer showing that three named alternatives addressed a buying condition while the client did not enter the answer.

The gap also protects the agency’s commercial position. The retainer is not justified by a promise of placement. It is justified by a recurring cycle of finding missing evidence, improving relevant public surfaces, and checking whether the answer changes.

## Make UK legal tech prompts sound like real procurement

Legal tech reporting improves when the prompt set reflects how UK buyers evaluate software, rather than how a generic keyword list describes the category. Test jurisdiction, firm size, practice area, security requirements, implementation burden, and integrations when those factors affect selection.

### Three prompt groups belong in the baseline

**Jurisdiction and compliance prompts** test whether the client is associated with UK practice and data expectations. Examples include “which legal case-management systems suit UK solicitors’ firms?” and “what should a UK law firm check before choosing cloud case-management software?”

**Operational prompts** test fit for the work itself. Examples include “compare legal intake tools for a UK conveyancing team” and “which software helps a small UK firm manage matters, deadlines, documents, and client communication?”

**Procurement prompts** test the information needed for a shortlist. Examples include “what does implementation involve for a 40-person UK law firm?” and “which legal software vendors publish clear pricing or explain their pricing drivers?”

Each prompt needs a reason for inclusion. If a legal tech client does not serve conveyancing, remove conveyancing questions. If the product sells only to enterprise legal departments, do not let small-firm prompts shape the main share-of-voice view.

Agencies serving other UK software categories can apply the same reporting discipline without copying the legal vocabulary. A healthcare SaaS provider may need NHS procurement and clinical governance questions. A logistics technology provider may need fleet size, depot coverage, and UK carrier integrations. The report earns trust when the prompt set resembles a sales call.

While you read this

Somewhere right now, ChatGPT is recommending a vendor in your category.

Run a free check and see whether it names you or a competitor. No credit card.

[Check your AI search visibility](https://www.citedintel.com/free-geo-tool)

## White-label reports should hide the logo, not the proof

White-label reporting works when the agency owns the interpretation and the evidence remains inspectable. Removing product branding is reasonable; removing prompts, dates, sources, and caveats makes the document harder for a client to trust.

A practical agency report can use three layers:

- **Executive page:** movement, commercial implication, risk, and next decision.
- **Account page:** prompt groups, rival share, changes, and recommended work.
- **Evidence appendix:** answer excerpts, source URLs, dates, engine records, and methodology.

Use the agency’s colors and terminology on the first two layers. Keep the appendix plain enough for a client’s product, legal, or sales team to audit. A white-label document should not make an assistant answer look like a guaranteed media placement.

Citedintel supports agency reporting with a demo-led Agency Pro plan, including five team seats and four client projects, with pooled audits and capacity scoped with you as usage grows. Enterprise and Whitelabel & Custom arrangements cover custom limits, unlimited seats, and brand voice requirements. Present those details only when a client asks how the agency produces the report, not as a substitute for the findings.

For an agency lead, the useful handoff is an editable client-facing summary plus an internal evidence file. The account team can discuss priorities without exposing internal notes, while the strategist retains the source trail needed for the next audit.

## Write the limitations into the methodology

AI answers vary by engine, query formulation, location, user context, retrieval sources, and day. A monthly report should state that results are observations from a defined audit set, not stable rankings or a promise of inclusion.

OpenAI’s current Search guidance says ChatGPT can reformulate questions and retrieve information through search partners. Stanford research published in February 2026 also found that web-search performance across 15 large language models varied with query wording, source priorities, and how retrieved material was combined. Link those facts in the methodology so the client understands why repeated audits matter.

Do not promise that a page, review, or digital PR placement will produce a recommendation. Google’s July 2026 guidance says best practices do not guarantee that a page will be crawled, indexed, or served, and the same caution applies to AI answer inclusion.

A second interpretation point matters for commercial reporting: a mention does not prove trust or conversion. Pew reported in June 2026 that six in ten US adults had read AI-generated search summaries, while 6% said they trusted those summaries “a lot”; the figures describe US adults rather than UK software buyers, so use them as context rather than as a UK forecast.

For B2B software, source accuracy deserves attention because inaccurate advice can weaken a buying decision. A February 2026 Forrester report found that 19% of B2B buyers using generative AI applications felt less confident because of unreliable information. Show the source behind a recommendation, then check that the source still states the client’s current product facts.

## Copy this report section into your next deck

The following skeleton gives an agency a repeatable section without turning every client report into a generic dashboard. Replace the bracketed fields, retain the evidence links, and remove any metric that lacks a stable prompt base.

> **AI search optimization report: [Client] | [Month Year]**
>
> **Market and scope:** United Kingdom | Location: [setting] | Language: [language] | Audit dates: [dates]
>
> **Business question:** [What should a UK buyer be able to find or understand about the client this month?]
>
> **Prompt coverage:** [number] prompts across [category], [requirements], [comparison], [switching], [pricing], and [implementation].
>
> **Headline movement:** Mentions [this month] versus [previous audit]; recommendations [this month] versus [previous audit]; citations [this month] versus [previous audit]; recommendation share against [rival set] [this month] versus [previous audit].
>
> **What changed:** [One observed movement, with prompt group, engine, date, and location.]
>
> **Where the client entered answers:** [Prompt, answer excerpt, position, and cited source.]
>
> **Where the client was missing:** [Prompt, rivals shown, buyer requirement, and likely evidence gap.]
>
> **Source quality:** [Current, incomplete, outdated, incorrect, or absent], with URLs for each finding.
>
> **Recommended work:** [Asset or source action] | Owner: [name] | Publish or outreach date: [date] | Recheck: [date].
>
> **Method note:** Results can vary by engine, query wording, location, user context, retrieval sources, and audit date. The report records observed answer behavior and does not represent a ranking promise.

The skeleton gives the client a direct line from question to evidence to work. It also gives the agency a defensible record when an answer changes before the next meeting.

## Try this today: make a dated UK evidence sheet

You can create a useful baseline with a small dated prompt set and a simple comparison sheet. The result should reveal whether the client’s absence is broad or limited to a specific UK requirement.

1. **Choose one vertical:** use legal tech for this example, then name the client’s real buyer and firm type.
2. **Write six prompts:** category, UK requirement, operational need, competitor comparison, implementation, and pricing.
3. **Run each prompt twice:** record the engine, date, UK location setting, full answer, client mention, recommendation, and cited URLs.
4. **Name three alternatives:** take the names from current lost-deal notes or the client’s sales team, not from a generic software list.
5. **Mark the answer role:** use “mentioned”, “recommended”, “compared”, “cited”, or “absent”. Do not merge the labels.
6. **Circle one gap:** choose the missing requirement that appears in a rival’s explanation but not in the client’s public evidence.
7. **Assign one fix:** specify the page, source, review, or public reference to create or update, with an owner and recheck date.

For a UK legal tech client, the sheet might test “case management for a small UK solicitors’ firm”, “legal software with Microsoft integration”, “secure document management for UK law firms”, “alternatives to [named vendor]”, “implementation for a 40-person firm”, and “legal software pricing in the UK”.

Citedintel lets an agency turn that same evidence sheet into recurring buyer-prompt audits, diagnosis, fix briefs, and client-ready reporting, with the [full workflow described here](https://www.citedintel.com/why-cited).

## Use the findings to define the next month’s work

The strongest monthly recommendation has three parts: the answer gap, the public evidence needed to address it, and the next observation date. That structure moves the conversation away from an arbitrary page quota.

An account lead can present the work as a sequence: repair the UK service page, publish a comparison that states fit and limitations, correct a third-party profile, then recheck the same prompt group. The client can approve each action because the report names the buyer question it serves.

Connect the findings to wider buyer behavior without claiming revenue the audit cannot prove. McKinsey reported in May 2026 that B2B buyers use an average of ten channels and that inconsistent information is a leading driver of supplier switching. If an assistant says one thing, the website says another, and a review profile is outdated, the report should flag the inconsistency across those buying surfaces.

Keep pipeline attribution separate. A client may ask whether a recommendation produced a sale, but the report can only claim observed answer presence unless analytics, buyer research, and CRM evidence connect the answer to a real opportunity. Agencies that keep that boundary clear are better placed to retain trust when the numbers move in different directions.

For the UK market hub, use the [United Kingdom AI search optimization guide](https://www.citedintel.com/ai-seo/united-kingdom) when you need regional context for the wider program.

## Make the report a decision record

A good agency report tells a client what appeared, why the result matters, what evidence supported it, and which UK buyer question deserves work next. It distinguishes a name in an answer from a recommendation that a buyer could act on.

My second strong view is that agencies should stop selling AI SEO as a promise of position. Sell a measured operating process instead: fixed questions, dated observations, competitive context, source review, publishable work, and a recheck that can confirm or challenge the previous finding.

When a client asks, “Are we showing up in ChatGPT?”, the answer should open a wider conversation about AI search engine optimization, buyer requirements, and proof. A report built that way gives the client a reason to renew because each month produces a sharper decision, not another dashboard.

Citedintel offers an interactive [demo for agencies planning this reporting workflow](https://www.citedintel.com/demo); the link is a product demonstration, not independent research or a promise of client results.

*Parth Sesodia is Founder, Cited.*