How to Choose an AI Search Optimisation Agency in the UK

Choose an AI search optimisation agency by asking it to show the path from raw answer evidence to implemented business change.

A credible agency will expose its prompt sample, cited URLs, technical findings, content decisions, authority plan and measurement limits. It will explain where the work overlaps with SEO and charge only for additional labour. It will never guarantee that ChatGPT, Gemini, Perplexity or Google will recommend you.

An unconvincing agency sells a proprietary score, an llms.txt file and a monthly screenshot deck.

Checked: 14 July 2026 · Written for UK businesses comparing AEO, GEO and AI-search providers

The 12-test agency scorecard

Score each provider 0 absent, 1 claimed, 2 evidenced. A strong shortlist should reach 19 out of 24 without a zero on raw evidence, implementation, SEO overlap or ownership.

Ask for evidence during the selection process. A methodology should survive daylight before the contract begins.

First, understand what you are hiring

Providers use several names:

• answer engine optimisation (AEO);
• generative engine optimisation (GEO);
• AI search optimisation;
• LLM visibility;
• AI SEO.

The useful work usually combines:

1. technical search eligibility;
2. entity and business-fact clarity;
3. answer and citation sampling;
4. commercially useful content;
5. structured data matching visible facts;
6. trusted external corroboration;
7. search, referral and conversion measurement.

Names matter less than scope. Ask the agency to show which of those it implements and which it only audits.

Test 1: does it define the market before running prompts?

A baseline needs boundaries:

• countries and languages;
• products or services;
• ideal customer and buying stage;
• local, national or international intent;
• named AI products and modes;
• competitors and alternatives;
• the date and environment of the sample.

A hundred prompts across every service can be less useful than thirty questions attached to one valuable customer journey.

The agency should build prompts from Search Console, sales calls, customer research, site search and category behaviour—not only from an AI tool generating variations of the keyword.

Test 2: can you inspect the raw answers?

Any “share of model”, “AI rank” or visibility score should link back to:

• exact prompt;
• engine and mode;
• run date;
• answer evidence;
• brands mentioned and their role;
• cited URLs;
• factual accuracy;
• method for repeat runs.

The score may be a useful summary. It is not the evidence.

Ask how the method handles answer variability. Strong answers discuss controlled wording, repeat sampling, prompt groups and visible uncertainty. Weak answers call one run a ranking.

Our guide to measuring AI search visibility gives a transparent five-layer model you can use to audit a vendor's report.

Test 3: does it inspect accuracy, not only presence?

A brand can appear for the wrong service, wrong location or another company with the same name. A citation can point to an outdated review or a weak third-party page.

Require classification of:

• accurate recommendation;
• accurate comparison option;
• citation without recommendation;
• passing mention;
• material factual error;
• mistaken identity;
• unsuitable recommendation.

Inaccurate visibility creates a correction task, not a success metric.

Test 4: does the technical audit match how systems actually work?

For Google AI Overviews and AI Mode, Google's official guidance says standard SEO best practices remain relevant, a supporting page needs to be indexed and snippet-eligible, and no special AI schema or text file is required.

For ChatGPT search, OpenAI advises publishers not to block OAI-SearchBot if they want content included in summaries and snippets in its publisher FAQ.

A proper technical review therefore covers:

• crawler and CDN access;
• indexability and canonicalisation;
• rendered text and internal discovery;
• duplicate and conflicting pages;
• organisation, person, service and product structure;
• structured data matching visible content;
• business profiles and merchant information where relevant;
• referral and conversion tracking.

Installing llms.txt without checking the indexable website is a distraction.

Test 5: can it explain the overlap with SEO?

Ask for a scope table:

Populate it for technical fixes, content briefs, schema, internal linking, digital PR, prompt monitoring, citation analysis and reporting.

Two agencies editing the same page from separate keyword lists create conflict. A good AI-search provider can work inside an existing SEO programme or show why responsibilities should be consolidated.

See AEO pricing UK for the differences between audits, monitoring and monthly implementation. Pay for new research and implementation, not renamed foundations.

Test 6: does content begin with buyer questions and evidence?

The agency should inspect existing commercial pages before proposing a publishing programme.

High-value gaps often include:

• who the offer is for and not for;
• costs and cost drivers;
• comparisons and alternatives;
• implementation and migration risk;
• evidence, methodology and limitations;
• location, availability and service boundaries;
• original data or experience;
• clear authorship and refresh dates.

Ask how each new page avoids cannibalising an existing one. “We will publish 20 AI-optimised articles a month” is a production promise without a search architecture.

Test 7: is there an authority and source strategy?

Answer engines often use a wider source set than the brand website. The agency should identify:

• industry bodies and standards;
• authoritative documentation;
• credible directories and profiles;
• customer and expert communities;
• comparison and review sources;
• publications and datasets;
• partners, events and experts;
• sources currently cited in sampled answers.

Then separate what can be corrected, earned, contributed to or merely monitored.

Schema cannot create external consensus. A source strategy might involve original research, expert commentary, partner documentation, accurate profiles, customer proof or digital PR. Buying bulk mentions on irrelevant sites is not corroboration.

Test 8: what will be implemented in 90 days?

A practical sequence:

Days 1–30

• define scope and baseline;
• resolve crawler, indexing and material entity conflicts;
• protect measurement access;
• identify the five highest-value content or source gaps.

Days 31–60

• improve priority service, product and comparison pages;
• strengthen internal links and matching structured data;
• correct owned profiles and obvious external inconsistencies;
• produce one defensible proof or data asset where justified.

Days 61–90

• repeat the controlled prompt sample;
• review search and referral changes;
• promote useful assets to relevant sources;
• update the roadmap from evidence;
• document what did not move and why.

Ask who performs each task. An audit recommendation is not implementation.

Test 9: does measurement admit platform limits?

Google's Generative AI performance report is available to a limited set of sites. Ask whether your property has access and which metrics the provider can actually report. Ordinary Web totals do not isolate AI activity. AI answers vary, and consent, apps and copied links leave gaps in referral analytics.

The agency should report:

• technical eligibility;
• controlled sampled visibility by prompt group;
• cited sources and factual accuracy;
• Search Console page-query trends without false AI attribution;
• identifiable AI referrals;
• qualified business actions and stated tracking gaps;
• work shipped and the decision it caused.

If the vendor promises exact market-wide AI share, ask where the complete population of answers comes from.

Test 10: does it know when not to recommend AEO?

Pause the specialist programme when:

• the website does not clearly explain the offer;
• core pages cannot be indexed;
• product or location facts conflict;
• the company lacks proof buyers need;
• nobody can implement recommendations;
• sales cannot identify a qualified enquiry;
• ordinary search demand and customer research have not been understood.

The cheapest path may be fixing the source website and measurement first. An honest provider should say so.

Test 11: are commercial claims careful?

Reject:

• guaranteed citations or recommendations;
• claims of a private relationship with answer engines;
• promises that one file makes a business “AI-ready”;
• fixed percentage lifts without a baseline;
• case studies that omit dates, scope and method;
• screenshots with no prompt or source context.

Ask what the agency guarantees. Good answers include work quality, documented checks, access, communication and delivery—not third-party selection.

Test 12: can you leave with the work?

You should retain:

• prompt set and methodology;
• raw answer and citation exports;
• technical findings;
• content, research and briefs you bought;
• account and analytics access;
• schema and site changes;
• source and outreach records appropriate to the contract;
• reports, refresh dates and documentation.

Avoid proprietary dashboards as the only record. Exportable evidence keeps the baseline useful after the subscription ends.

Questions for the pitch

1. Show us the raw data behind a client visibility report.
2. Which part of our current SEO work would you not duplicate?
3. How do you test answer variability?
4. Which AI visibility claims can you not make?
5. What do Google and OpenAI officially require today?
6. Which five changes would you inspect before recommending new content?
7. How do you assess third-party sources without buying low-quality mentions?
8. Who implements technical and content recommendations?
9. How will reporting connect to qualified enquiries?
10. What can we export when we leave?

Frequently asked questions

What does an AI search optimisation agency do?

It audits technical access, entity clarity, answer visibility, citations, commercial content and external corroboration, then implements improvements and measures a controlled sample alongside search, referral and commercial data.

How much should an AEO agency cost?

Compare the work before comparing the fee. An audit, monitoring subscription and managed implementation programme are different purchases. Ampliflow's published SEO and AI-search starting point is £497/month, subject to agreed scope; it is not a one-off audit price. Our AEO pricing guide explains the exclusions and questions to put into a brief.

Can an agency get my business into ChatGPT?

No agency controls selection. It can make the business crawlable, clear, evidenced and well corroborated and can measure sampled visibility. It cannot guarantee inclusion.

Is llms.txt a sign of a capable agency?

No. It may be a small experimental deliverable, but Google says no special AI text file is required for its AI features. Judge the agency on the website, evidence, sources and implementation.

Should my SEO agency handle AEO?

It can if it has a transparent answer-sampling method, understands citations and entities, and integrates the work with technical SEO, content and authority. The label on the supplier matters less than the capability and ownership model.

Choose the agency that makes uncertainty manageable

The category is new enough for overclaiming and mature enough for disciplined work. Buy transparent evidence, clear implementation and commercial judgement. Leave the magic scores to the pitch deck.

Related: SEO agency cost UK · SEO agency red flags · AI visibility audit deliverables · From SEO to AEO · Measure AI search visibility · AEO pricing UK