In 2026, every marketing agency claims to be AI-powered. Almost none of them will tell you what that means operationally.
That is the actual problem you are trying to solve. You are not trying to find the agency with the best AI page on their website. You are trying to work out which of them has changed how the work gets done, and which of them added the word “AI” to a deck they were already using in 2023.
This page gives you the questions to ask rather than a ranked list. There are already at least seven ranked lists competing for this query, and every one of them will tell you a different agency is number one. A framework travels better than a leaderboard, because it survives the next twelve agencies you talk to.
GrowthRocks is a growth marketing agency and we use AI in delivery. Specifically: clustering and analysis at scale, forecasting, and optimising client properties for citation in AI search surfaces. We are not selling a proprietary model. Being precise about that is the whole point of this article, so it would be strange not to apply it to ourselves.
The claim is not binary. There are four tiers, and they represent enormously different amounts of actual capability. Most agencies marketing themselves as AI-powered are at tier one.
| Tier | What it means | Is it a differentiator? | How to verify |
|---|---|---|---|
| 1. AI-assisted production | The team uses ChatGPT, Claude or similar for drafting, research and summarising | No. This is table stakes in 2026 and everyone does it | Ask what percentage of deliverables involve AI drafting and whether they disclose it |
| 2. AI-assisted analysis | Keyword and audience clustering at scale, forecasting, anomaly detection, creative testing analysis | Mildly. Depends on whether it changes decisions | Ask for a decision they made from an AI-driven analysis that they would not have made otherwise |
| 3. Custom models and agents | They build pipelines, agents or fine-tuned models against client data | Yes, genuinely | Ask what it runs on, who owns it, and what happens to it when you leave |
| 4. AI search optimisation (GEO) | They optimise your brand to be cited by AI systems | Yes, and it is a separate discipline | Ask for their measurement methodology before anything else |
The useful question is not “do you use AI”. Everyone will say yes. The useful question is “which tier are you at, and what specifically do you do at that tier”. An agency operating honestly at tier two beats an agency claiming tier four vaguely, every time.
One caution about tier three. Building custom tooling is impressive but it is not automatically in your interest. Ask who owns the prompts, the agents and the fine-tuned artefacts when the engagement ends. If the answer is that they own everything, you are renting a capability rather than building one, and you should price that accordingly.
These three terms get used as synonyms and they describe different work. Sorting them out is most of the battle.
AI marketing agency is the umbrella term. It usually means an agency that uses AI in delivery across channels, and sometimes means an agency that specialises in marketing AI products. Those are two entirely different propositions sharing one label, which is why the category is so muddled. Ask which one they mean.
AI SEO agency means using AI to do SEO work faster or better: clustering keywords at scale, generating and editing content with AI assistance, automating technical audits, analysing SERP patterns. The output is still traditional search rankings. The AI is in the production method.
GEO, or generative engine optimisation means optimising so that AI systems cite you. The output is not a ranking, it is a citation inside an AI Overview, an AI Mode response, or a ChatGPT answer. This is a genuinely different discipline with different tactics and, critically, different measurement.
The distinction matters commercially. If your traffic is falling because AI Overviews are answering your queries above your listing, an AI SEO agency that makes your existing content production faster does not solve your problem. You need the third thing. Many agencies sell the second and describe it as the third.
If you want the underlying mechanics of how AI search surfaces work, we have covered Google AI Mode in detail separately.
For each question, what a good answer sounds like and what a bad one sounds like.
1. Which of the four tiers do you operate at, and what specifically do you do there? Good: They pick a tier and describe the work concretely, including what they do not do. Bad: “All of them.” Nobody does all four well.
2. What proportion of client deliverables involve AI generation, and do you disclose it? Good: A number, and a stated disclosure policy. Bad: Discomfort with the question, or “our process is proprietary”. You are paying for the output. You are entitled to know how it is made.
3. How do you handle the risk of AI-detectable content? Good: They talk about editorial review, subject matter input, and original data, and they note that detection tools are unreliable in both directions. Bad: They claim their content is undetectable. That is a production claim, not a quality claim, and it tells you what they optimise for.
4. What is your GEO measurement methodology? Good: Named tools, a defined query set, a stated sampling frequency, and an acknowledgement that this measurement is immature. Bad: “We track AI visibility” with no method behind it. Ask this before anything else if AI search is why you are hiring.
5. Can you show me a before and after on AI citations for a real client? Good: A specific account, a query set, and citation counts at two points in time. Bad: A screenshot of one AI Overview mentioning a client, with no baseline.
6. Whose models, and whose data? Good: Clarity on which providers, whether client data touches training, and what their data processing terms say. Bad: Vagueness. This is a contractual and compliance question, not a technical curiosity.
7. What happens to the prompts, agents and custom tooling when we leave? Good: A defined handover, or an honest statement that it stays with them and pricing that reflects that. Bad: They have not thought about it.
8. How do you validate AI output for factual accuracy? Good: A named review step with a named human owner, and an example of something that was caught. Bad: “Our tools are very accurate.”
9. What does your team do that the tools cannot? Good: Strategy, positioning judgment, subject matter access, original research, taste. Something specific. Bad: A list of tools. If the answer to what humans add is another product name, you are paying agency margin for software you could license yourself.
10. Has your pricing changed as AI reduced your delivery cost, and if not, why not? Good: An honest answer, which might be that rates held but output volume or quality rose. That is defensible. Bad: Deflection. This question is uncomfortable on purpose. The answer tells you whether they see AI as a client benefit or a margin benefit.
Ranges observed across European and US engagements in 2026. Pricing varies heavily by tier, which is exactly why the tier question comes first.
| Engagement | Typical monthly range (USD) |
|---|---|
| Tier 1 to 2 content and SEO retainer | 3,000 to 8,000 |
| Tier 2 full-funnel growth retainer | 8,000 to 20,000 |
| Tier 3 custom model or agent build | 15,000 to 60,000 project, then retainer |
| Tier 4 GEO programme (standalone) | 4,000 to 15,000 |
| AI search visibility audit (one-off) | 2,500 to 10,000 |
A note on the pricing paradox. AI has meaningfully reduced the cost of producing marketing deliverables. If an agency’s rate card has not moved since 2023, ask what they did with the difference. There are good answers, such as reinvesting in strategy time, original research, or more experiments per month. There are also bad ones. Either way you should hear the answer before you sign, because it tells you whether you are buying leverage or subsidising margin.
Grouped by tier rather than ranked, since the tiers are not comparable. Verify current positioning and pricing directly.
Coalition Technologies. Positions explicitly around ranking in ChatGPT and AI Overviews, which is the correct framing for GEO work rather than AI-assisted SEO.
Onely. Technical SEO specialists who have moved into AI search analysis. Strong fit where the constraint is crawlability, rendering and technical foundations rather than content volume.
Searchbloom. Markets around AEO and GEO alongside traditional SEO in the US market.
NoGood. Has built proprietary tooling around answer engine optimisation and AI discoverability, and works with AI-first brands. One of the few agencies with a defensible tier three claim.
Level Agency. Performance-oriented digital marketing with AI integrated into optimisation and testing.
Omniscient Digital. Content and SEO focused, publishes original research, and writes critically about the AI agency category itself, which is usually a good sign.
Thrive Agency. Broad digital marketing with an AI SEO service line. Full-service breadth rather than AI depth.
Major Tom. Full-service digital agency using AI for paid media optimisation, bid management and content generation across a large client roster.
GrowthRocks. We operate at tier two with tier four work as a specialism. Concretely: clustering and forecasting in our analysis workflows, and GEO programmes where we baseline citation counts across AI Overviews, AI Mode, ChatGPT, Gemini, Perplexity, Copilot and Grok, then work to move them. We do not build custom models, we do not sell AI-generated content volume, and we disclose AI involvement in deliverables. If you need tier three custom model work, we are not the right call.
This is where most engagements become unfalsifiable. Insist on these before you start.
Citation frequency across AI surfaces. Count how often your domain is cited in AI Overviews, AI Mode, and the major assistants, against a fixed query set. Fixed is the important word. A query set that changes between measurements produces numbers that cannot be compared. Ahrefs Brand Radar handles this, as do several dedicated GEO tools.
Share of voice against named competitors. Your citation count in isolation means little. Measured against three named competitors on the same query set, it means something.
Branded search volume. A lagging indicator, but a real one. If AI systems are recommending you, some proportion of people will search your brand afterwards. Watch it in Search Console over quarters, not weeks.
Referral traffic from AI platforms. Small in absolute terms for most sites, and worth segmenting anyway because it converts differently.
What not to accept: a single AI Overview screenshot, an “AI visibility score” with no published method, or impressions presented as citations.
One honest caveat. This measurement discipline is roughly eighteen months old. The tools disagree with each other, results vary by geography and session, and anyone presenting AI visibility numbers to three decimal places is overselling the precision available. Directionally useful, not precise.
A worked example, since we are asking you to demand evidence. Measured across all major AI platforms in July 2026, growthrocks.com held 94 response-level citations: 38 in Google AI Mode, 17 in Gemini, 14 in AI Overviews, 10 in Grok, 8 in Perplexity, 4 in Copilot and 3 in ChatGPT. For a domain with roughly 1,800 referring domains and a decade of content, that is a low footprint, and it is the baseline we are working against ourselves. Any agency asking you to trust their AI expertise should be willing to show you their own numbers.
What is an AI marketing agency? An agency that uses artificial intelligence in how it delivers marketing work. In practice this spans four levels: AI-assisted content production, AI-assisted analysis, custom models and agents built on client data, and optimising clients to be cited by AI search systems. Most agencies using the label operate at the first level.
What is the difference between AI SEO and GEO? AI SEO means using AI to do search optimisation work more efficiently, with traditional rankings as the output. GEO means optimising so AI systems cite your brand in their generated answers. Different tactics, different measurement, frequently sold as the same thing.
How much does an AI SEO agency cost? Typically 3,000 to 8,000 USD per month for content and SEO retainers, 4,000 to 15,000 for standalone GEO programmes, and 8,000 to 20,000 for full-funnel growth work. The main variable is whether you are buying production efficiency or genuine AI search expertise.
Can an AI agency guarantee AI Overview placement? No. AI systems are non-deterministic, results vary by user and session, and no agency controls the output. Anyone guaranteeing placement is misrepresenting how these systems work.
Is AI-generated content bad for SEO? Not inherently. Google’s guidance targets low-value content produced primarily to manipulate rankings, regardless of how it was made. The determinant is whether the content is useful, accurate and original, not its provenance. In practice, AI-generated content fails more often because it is generic, and generic content was already failing before AI existed.
How do I know if an agency is really using AI or just claiming it? Ask them to name a decision they made from an AI-driven analysis that they would not have made otherwise. Tier one agencies cannot answer it, because using AI to draft faster does not change decisions.
Should I hire an AI SEO agency or train my existing team? Train your team if you have three or more people already doing SEO and content, since the tooling is accessible and the skills transfer. Hire externally if AI search visibility is a strategic priority and you have nobody who has done GEO measurement before, because that methodology is the part that is genuinely hard to self-teach right now.
💡 Not sure which one fits your needs? Get in touch – we’re happy to help you figure it out.
I write for GrowthRocks, one of the top growth hacking agencies. For some mysterious reason, I write on the internet yet I’m not a vegan, I don’t do yoga and I don’t drink smoothies.
Built an AI routine that checks who is emailing me, understands the business context, filters…
Traditional online communities were built for a different internet. Community 2.0 brings humans and AI…
MCP-first apps change how software is built and used. Discover why the UI is becoming…
A step-by-step methodology for measuring how often AI systems cite your brand, with the query…
In 2026, growth hacking is no longer about finding single-turn shortcuts; it has evolved into…
Accessing AI data is only half the challenge. Learn how to transition from static Markdown…