A Founder's Buyer's Guide. Six capabilities to evaluate, a scoring rubric, 40+ questions for discovery calls, and the red flags that should make you walk away.
Updated April 2026 · 12 min read · By Tom Parling, Founder, growthvibe
Since launching growthvibe at the end of February, I've been on more than 100 calls with prospective clients and another 50 with agencies, vendors, and practitioners in the AEO and GEO space. A lot of those client calls are in-house buyers trying to pick an AEO agency, firing off questions, trying to work out who's credible. The quality of those questions varies wildly. Some are sharp. Most aren't.
Better questions produce better answers. Better answers mean you hire the right agency. That's the whole point of what follows.
Use it to score growthvibe as much as anyone else. If we come out badly against the criteria below, hire someone else. I'd rather lose a pitch honestly than win one because the buyer asked the wrong questions.
Roughly half of Google searches already return an AI summary, and McKinsey expects that to pass 75% by 2028. Sixty percent of US searches now end without a click at all (SparkToro). Brands cited in Google AI Overviews earn 35% more organic clicks and 91% more paid clicks than those that aren't (Seer Interactive). When AI Overviews are present and a brand isn't cited, organic CTR drops 61% and paid CTR drops 68% (Seer Interactive).
Those numbers together tell you everything about the commercial stakes. AEO isn't a future channel. It's a live one, and the gap between cited brands and non-cited brands is widening faster than buyers are adjusting.
The agency you choose to work with in 2026 will either close that gap for you or make it worse. Picking well is worth the hour it takes to read this.
Three steps.
Shortlist three to five agencies that position themselves explicitly around AEO, GEO, or AI visibility. If their homepage leads with "SEO" and AEO is a subpage, they're an SEO agency that does AEO on the side, and you want a specialist.
Take the questions below into discovery calls and demos. Don't accept marketing answers. Ask for evidence, exported data, or reproducible examples. Write down how specific each agency's answer is, because specificity is the single best proxy for whether the capability is real.
Score against the "what good looks like" column. The right partner will satisfy most criteria across all six capabilities. Be suspicious of an agency that is strong on one or two and weak on the rest.
Use this as a one-page summary for internal stakeholders. Score each capability 1–5 based on the agency's answers to the full question bank underneath.
| Capability | What you're testing | Minimum bar to clear | Score 1–5 |
|---|---|---|---|
| 1. Measurement | Multi-engine tracking, share of voice, exportable raw data, AI bot analytics | At least 5 engines covered, weekly refresh, exportable prompt-and-response pairs | |
| 2. Entity and authority foundations | Entity perception audit, Wikidata and Knowledge Panel methodology, citation infrastructure | Can name the entity stack and ship a baseline entity perception audit as a standard deliverable | |
| 3. AI-optimised content and technical | Structural differences vs SEO, schema mapping, bot accessibility verification | Before/after examples that go beyond "added a FAQ", documented content-type-to-schema mapping | |
| 4. Commercial proof | Attribution model, LLM-referred cohort reporting, named case studies with measurable citation movement | Two named clients with dated citation-rate shifts and a stated attribution model | |
| 5. Team calibre and originality | Named operators, founder involvement cadence, dated original research | Named individuals on the account plus original published research within the last 90 days | |
| 6. Engagement model | Retainer structure, minimum term, pricing logic, conflict policy | Strategic retainer, 6-month minimum, scope-based pricing, disclosed conflict policy |
A strong AEO partner should score 4 or 5 on at least four capabilities. Weakness in Capability 1 (measurement) or Capability 2 (entity foundations) is disqualifying regardless of strength elsewhere. The other four can compensate for each other. Those two cannot.
When we pitch a brand, the first thing I ask is how they currently measure AI visibility. Of the last twelve discovery conversations, nine incumbents were measuring by taking screenshots of ChatGPT responses once a month. Two were using a single third-party tool with no comparative baseline. One was doing it seriously.
Measurement infrastructure is the foundation of every credible AEO programme. If the agency can't measure across the engines, they can't prove outcomes, and you will be paying for activity rather than results.
| Question to ask | What good looks like |
|---|---|
| Across which answer engines do you measure? | ChatGPT, Google AI Overviews, Perplexity, Gemini, and Claude at minimum. Copilot, DeepSeek, Grok, Meta AI, and Google AI Mode if you operate internationally. If they only measure ChatGPT, the stack is a year behind. |
| Do you quantify share of voice against competitors, or only measure presence? | Share of voice against named competitors, citation rate by prompt category, citation depth by page or entity, and sentiment of surrounding content. Presence alone tells you nothing about whether you're winning or losing. |
| How often is the data refreshed? | Weekly at minimum. AI answer outputs shift continuously. Monthly data is already stale when you see it. |
| Is the raw data exportable? | Yes, as prompt-and-response pairs. If you can't audit the underlying data, you can't verify the uplift claims later. |
| Question to ask | What good looks like |
|---|---|
| How are the tracked prompts selected? | Hybrid. They start with your commercial priority categories and layer in observed real-world prompt data. A static list of keywords ported over from SEO is the wrong answer. |
| Do they distinguish branded, competitor, and category-level prompts? | Yes, scored separately. Branded visibility is a floor. Category visibility is the growth opportunity and the one worth paying for. |
| How are new prompts added as the market evolves? | A repeatable process based on emergent prompt data in your category, not a list fixed at engagement start. If the answer involves asking you for more keywords quarterly, they're running an SEO workflow. |
| Question to ask | What good looks like |
|---|---|
| Can you tell me which AI bots are visiting my site, how often, and which pages? | Yes, via CDN integration (Cloudflare, Vercel, Fastly, CloudFront, Akamai), server log ingestion, or a CMS-level module. An agency that doesn't know which bots are crawling your site can't diagnose accessibility problems. |
| Can you classify bots by purpose? | They should separate AI training bots, AI answer retrieval bots, and classical search crawlers. Each implies different optimisation choices. |
| Can you measure human referral traffic from answer engines, with conversion? | Yes, with a stated, documented approach to attribution. The attribution will be imperfect. The approach to handling the imperfection is what you're buying. |
AEO is an entity game before it's a content game. If a partner can't explain how they position your brand as a recognised entity across the web, everything else they do sits on sand.
I've seen this go wrong most often where an agency leads with content production. They'll publish fifty AEO-optimised articles in a quarter, and the brand still won't be cited by AI engines, because the entity signals underneath the content aren't there. Content without entity infrastructure is fishing without a line.
| Question to ask | What good looks like |
|---|---|
| How do you audit how AI engines currently perceive our brand? | A baseline entity perception audit as a standard deliverable. Before any intervention, you should know what each major answer engine thinks your brand is, does, and competes with. |
| How do you define and reinforce our entity across the web? | Structured process covering Wikidata, Google Knowledge Panel, Organization schema, entity disambiguation, entity relationship mapping, and third-party citation reinforcement. If they can't name the components, the methodology isn't real. |
| How do you correct misperceptions? | Authority-source interventions (Wikipedia where merited, Wikidata, structured directories), PR-led citation building, schema correction, and direct platform interventions where available. |
| Question to ask | What good looks like |
|---|---|
| How do you identify which sources AI engines cite in our category? | From their measurement stack, not from industry lists. The cited sources differ materially by category, so a generic "here are the publications that matter" answer is a red flag. |
| How do you prioritise which citations to earn? | By observed citation frequency in your category, domain authority, entity relevance, and realistic feasibility. Not by list price. |
| Is this link building? | No, and if they describe it as link building they're running an SEO playbook. Citation authority is about source quality and entity reinforcement, not link equity. |
Most AEO claims live here, and most AEO programmes fail here. The question isn't whether the agency can produce content. It's whether they can explain the structural and semantic differences between content that gets cited by LLMs and content that doesn't.
A concrete test. Ask to see two versions of the same article they've optimised, before and after. If the "after" is the "before" with a FAQ section added, the methodology isn't real.
| Question to ask | What good looks like |
|---|---|
| What's structurally different about AEO content versus SEO content? | Answer-first formatting, entity-dense prose, explicit Q&A blocks, semantic chunking, fact-claim density, opinion-led passages, and internal linking by entity relationship rather than keyword. If the answer is "we add FAQs", it's not enough. |
| How do you identify which existing content to optimise? | Citation pattern analysis mapped against your inventory. They should know which of your pages are currently cited by AI engines before proposing new content. |
| Do you produce net new content, optimise existing, or both? | Both are valid. The order matters. A partner that proposes new content before auditing what you already have is skipping the cheap wins. |
| Question to ask | What good looks like |
|---|---|
| Which schema types do you implement, and how do you decide? | A documented mapping between content types and schema vocabularies (Organization, Product, Service, Article, FAQPage, HowTo, BreadcrumbList). Blanket schema application is a red flag, not a sign of thoroughness. |
| Do you implement llms.txt and agent-accessibility standards? | Yes, with a clear opinion on when it's useful. If they implement llms.txt on every client by default, they're box-ticking. |
| How do you verify AI bots can access the content? | Bot log analysis, crawl-path testing, rendered-content verification. An agency that optimises content without knowing whether the major AI bots can see it is guessing. |
| Question to ask | What good looks like |
|---|---|
| How do you select the questions? | Combination of observed answer engine prompts, sales and support question patterns, review mining, and category-level research. Not a generic "people also ask" dump. |
| Can you show me FAQ content you've published that is now cited by AI engines? | Yes, with the specific prompts that trigger the citation. This is one of the best proof points available in this category, and a competent agency will be eager to show it. |
AI visibility is a leading indicator. The commercial value of AEO is in what the visibility produces downstream. LLM-referred traffic converts at roughly 6x the rate of non-brand Google traffic in Webflow's data. The economics are meaningful if, and only if, the agency can connect the two ends.
Most agencies in this category can't. They'll report citation count, share of voice, and prompt coverage, and leave attribution for you to figure out. That's not enough in 2026.
| Question to ask | What good looks like |
|---|---|
| How do you attribute commercial outcomes to AEO work? | A documented model that accepts the limitations of AI referral tracking while still producing signal. Direct referral where available, branded search uplift as proxy, assisted conversion analysis, cohort measurement for major interventions. |
| Do you report LLM-referred traffic quality separately? | Yes, as its own cohort with conversion rate, session quality, and pipeline attribution where possible. |
| Can you benchmark us commercially, not just in visibility? | Share-of-voice trending, cited-competitor analysis, and third-party traffic benchmarks. "We'll track your visibility" isn't a commercial answer. |
| Question to ask | What good looks like |
|---|---|
| Can you show me clients where citation rate has moved measurably in the last six months? | Yes, with at least two named examples and an explanation of the specific interventions. Unnamed case studies are easier to fabricate. |
| What does a realistic twelve-month outcome look like for a brand in our sector and starting position? | A specific answer, not a range. "In markets like yours we typically see X move to Y, driven primarily by Z" is the right shape of answer. "It depends" is not. |
AEO is a senior discipline. The people doing the thinking matter more than the size of the team. This is the question I would push hardest on if I were buying.
The most common failure mode I see is senior operators running sales, junior operators running delivery. That's how agencies scale commercially and it's usually a bad deal for the client in a category as new as this one, where the strategy calls are the most valuable part of the engagement.
| Question to ask | What good looks like |
|---|---|
| Who specifically will run this account, named? | Individual names, with their AEO-specific background, published work, and time spent in the category. "A dedicated team" is a non-answer. |
| Will the founder or a senior strategist be involved, and at what cadence? | In mid-market and enterprise work, yes, at minimum at strategy review cadence. If the founder only shows up in sales, that's information. |
| Is technical implementation in-house or outsourced? | Either is fine if named. Opaque is not. You should know which agency and which individuals are writing the schema and shipping the changes. |
| Question to ask | What good looks like |
|---|---|
| What have you published originally about how AI engines cite, and when? | Dated pieces, ideally within the last ninety days. If their most recent original research is over a year old, the methodology isn't current. |
| Can you explain what has changed across the major answer engines in the last quarter and how it affects our programme? | Yes, in specifics. If the answer is generic, their monitoring isn't real. |
| Do you build your own tools or resell someone else's? | Valid either way, with one caveat. Agencies that have built their own AEO product are sometimes selling the product rather than optimising for your outcome. Ask whether you're a consultancy client or a tool user. |
AEO compounds. The right partner is engineered for long-term strategic engagement, and the commercial structure should reflect it.
| Question to ask | What good looks like |
|---|---|
| Retainer, project, or performance? | Strategic retainers. Projects can deliver artefacts. They rarely deliver compounding outcomes in this category. |
| What's the minimum engagement term? | Six months is the practical floor. Twelve is realistic for category-level repositioning. Anyone offering guaranteed outcomes in three months is selling a story. |
| How do you price? | Scope, market complexity, ambition. Not hours. If the pricing conversation feels like time-and-materials, the agency is commoditised. |
| Question to ask | What good looks like |
|---|---|
| Do you discount? | A specialist AEO partner shouldn't. Aggressive discounting is a signal of commoditisation, not a value add. I'd treat it as a disqualifier. |
| Do you work with direct competitors? | Reasonable disclosure, documented conflict policy. Specialists often serve multiple brands in a category, which is fine if ring-fenced, and you should know. |
| How do you define success in the first ninety days? | A specific plan: measurement baseline, two or three priority fixes, and the leading indicators you'll use to judge early progress. |
A few patterns I've seen enough times to take seriously. The checklist first, then the reasoning.
| Red flag | Why it disqualifies |
|---|---|
| "AEO is just SEO with schema" | Signals a rebadged SEO practice rather than the distinct measurement and entity methodology AEO requires. |
| Three-month outcome guarantees | Citation authority compounds. A quarterly guarantee either oversells or redefines "outcome" to mean something commercially empty. |
| Measurement by screenshot | No real measurement stack. If the demo shows screenshots of ChatGPT responses, walk. |
| Price-led pitches or early discounting | Specialists don't discount. A discount-first pitch signals commoditised delivery and thin methodology. |
| Junior operators on delivery, senior on sales | Strategy calls will be made by people without the category experience to make them well. |
| "We built our own AEO tool" with no disclosure of revenue model | Valid in principle, but if software revenue dominates, the agency optimises for tool usage rather than your outcome. Ask which revenue stream you'd sit in. |
| No original published research in the last 12 months | The engines change weekly. An agency that hasn't written publicly about it in a year isn't close enough to the work. |
| Generic prompt lists that look like keyword lists | Confirms the agency is porting an SEO workflow into AEO vocabulary without changing the underlying methodology. |
"AEO is just SEO with schema." It isn't. The measurement infrastructure, content methodology, and authority model are different enough that treating them as the same produces shallow work. An agency that opens with this line is telling you where their practice actually sits.
Three-month outcome guarantees. AEO compounds. Nobody can responsibly guarantee material citation authority shifts in a quarter. Anyone who does is either overselling or has redefined "outcomes" to mean something that isn't worth paying for.
Measurement by screenshot. If the measurement demo involves screenshots of ChatGPT responses, the stack doesn't exist. Walk.
Price-led pitches. If the first thing an agency offers is a discount, or a "let's find a package that works for your budget" reframe, the work isn't specialised enough to command its price. Pay attention to what this tells you.
Junior on delivery, senior on sales. Ask, directly, who will run the account. If the answer is evasive, assume the strategy calls will be made by people without the experience to make them well.
"We built our own AEO tool." Valid in principle. But an agency that's a software vendor by revenue model is optimising for tool usage, not for your outcome. Ask which revenue stream the account sits in.
Run discovery calls with three shortlisted agencies. Use the questions above. Don't compare the content of the answers. Compare how specific they are. The agency that responds with exportable data, dated original work, named individuals, and specific numbers is the one worth taking to a formal proposal.
Before any of it, baseline your AI visibility independently. If you know where you currently stand across the major answer engines before you start talking to agencies, you can score their diagnostic responses against a known truth. Any agency whose early diagnosis contradicts the baseline either has better data than you or doesn't know what they're looking at. Both are useful information.
The growthvibe AI Visibility Report is free and covers eight answer engines (ChatGPT, Google AI Overviews, Perplexity, Gemini, Claude, Copilot, DeepSeek, and Grok). Use it, or use a competitor's version of the same thing. The point isn't which tool. The point is that you shouldn't walk into agency conversations without a baseline.
Independent citation benchmark for your brand across the major answer engines. No sales call required.
Run the report →Functionally, almost always yes. AEO and GEO describe the same discipline with different origins. UK and European agencies lean AEO, US agencies lean GEO. If an agency makes a significant definitional point about one versus the other, ask what the substantive methodological difference is. Usually there isn't one.
Sometimes. Take the questions in this guide into the conversation. If they can answer the first three in specifics, add AEO to the engagement. If they can't answer any of them, hire a specialist.
Strategic retainers with specialist UK agencies start around £5k/month and scale based on scope and market complexity. An agency offering full AEO at £1k/month is offering something else. Aggressive discounting is a signal.
Six months to baseline and execute the first round of interventions. Twelve months to see material share-of-voice movement. Three-month sprints can deliver artefacts but not compounding results.
No. They overlap, they share foundations, and they'll converge commercially over time. Right now they are separate disciplines with different stacks and different methodologies. Run both. Measure both. Treat them as complements.
Organization and Person schema for entity grounding, Article and FAQPage for content extractability, Product and Service for commercial pages, HowTo where the content genuinely warrants it, and BreadcrumbList for site structure. The priority order is entity schema first (Organization, Person), then content schema (Article, FAQPage), then commercial schema (Product, Service). Blanket schema application is a red flag, not thoroughness. The right question isn't "which types do you add?" but "which types map to which content, and why?"
ChatGPT, Google AI Overviews, Perplexity, Gemini, and Claude as the baseline for any UK or US brand. Copilot, DeepSeek, Grok, Meta AI, and Google AI Mode as the second tier, weighted by where your audience actually researches. In B2B sectors, ChatGPT and Perplexity dominate consideration-stage prompts. In consumer search, Google AI Overviews and Gemini carry more weight. An agency that only tracks one engine, or weights them equally by default without sector-specific reasoning, is guessing.
Run an independent audit across the major answer engines before any discovery call. Score your presence, share of voice against named competitors, and citation depth on your top 20–50 commercial prompts. You can do this manually, through a third-party tool, or via the growthvibe AI Visibility Report, which runs across eight engines and is free. The point is to walk into agency conversations with a known truth. Any incumbent whose initial diagnosis contradicts the baseline either has better data than you or doesn't know what they're looking at. Both are useful.
Yes. growthvibe is UK-headquartered and UK-specialist, but the measurement stack covers the international engines (Copilot, DeepSeek, Grok, Meta AI, Google AI Mode in addition to the core five), and we work with brands in the US and Europe where there's a fit. The methodology is the same. The cited sources differ by market.
If you take nothing else from this, take the measurement test. The first thing any AEO agency should be able to do is show you, live, how they measure visibility across the major answer engines for a brand in your sector. If they can't, nothing else they say matters.
Tell us about your business and we'll come back with an honest view of your AI visibility, plus what we'd prioritise first.
No obligation. We'll respond within 24 hours with an initial assessment.