AI SEO Services: What They Include and How to Vet an Agency

“AI SEO services” now means at least four different things depending on who is selling. Some agencies mean using AI to write more blog posts. Some mean adding schema. A few mean the actual job, which is getting your brand named inside the answers your buyers read. This is the buy-side version: what belongs in scope, what does not, and the questions that separate the two in one call.

We sell this service, so read the vetting section adversarially and apply it to us too. Every question below is one we answer on our own sales calls.

Key takeaways

• A real AI SEO scope has six workstreams. If third-party mentions, original data, and citation measurement are missing, you are buying a content retainer with a new label.
• The single best filter is the agency’s own AI visibility. Ask ChatGPT and Perplexity who the best AI SEO agency is. An agency that cannot get itself cited will not get you cited.
• Using AI to produce content is not AI SEO, and it can work against you. Ahrefs’ study of 331,000 pages found no AI-content penalty, but low and moderate AI pages earned two to three times the impressions of high-AI pages.
• Demand a measurement method before you demand a price. “Citations went up” with no fixed prompt set and no baseline is unfalsifiable, and unfalsifiable reporting is the category’s biggest risk.
• Expect the payoff in booked conversations, not sessions. AI citations convert to clicks at a poor ratio by design, so a program measured purely on traffic will be cancelled before it pays.

What is actually in scope

Workstream	What it means	Why it matters	Commonly missing?
AI crawler access	Per-bot robots policy, render speed, static HTML, log monitoring	An engine cannot cite what it cannot fetch in time	Sometimes
Answer-first content	Rewriting money pages and posts so passages can be lifted whole	This is the stage citations are actually won at	No
Entity foundation	Consistent name, description, and facts across LinkedIn, Crunchbase, review sites, your own site	Engines skip attributions they are not confident in	Often
Original data	Publishing benchmarks or research only you can produce	Inverts the game: others cite you instead of you chasing mentions	Usually
Third-party mentions	Listicle inclusion, community presence, earned coverage	Most citations come from sources you do not own	Usually
Measurement	Fixed prompt set, citation share by engine, AI referral channel, self-reported attribution	Without it the program is unfalsifiable	Usually

The first two are table stakes and nearly every vendor does them. The last three are where programs succeed or quietly fail, and they are the ones to interrogate hardest, because they are slow, unglamorous, and easy to leave out of a proposal without anyone noticing for two quarters.

What is not AI SEO

Using AI to write more content. This is the most common substitution and the most expensive. It is a production-efficiency choice, not a visibility strategy, and the evidence says volume is not the lever. Ahrefs’ analysis of 331,000 pages found no penalty for AI-assisted content, but pages with low to moderate AI involvement earned two to three times the impressions of heavily AI-generated ones. Fast content that says nothing new gets neither ranked nor cited.

A schema package. Structured data is hygiene. The 2026 evidence does not support markup as a citation driver, and Google removed FAQ rich results entirely on 2026-05-07, so FAQPage markup no longer buys a visual result either. If schema is the headline deliverable, the proposal is thin.

Publishing llms.txt. It costs nothing and there is no evidence it earns citations. Reasonable to have, impossible to justify as a strategy.

A dashboard. Several vendors sell AI visibility tracking as the service. Monitoring is a component, not a program. Know whether you are buying work or a report.

AI SEO vs a traditional SEO retainer

The overlap is larger than either side admits, which is why the same work sometimes gets sold twice. The real differences are these:

• Unit of optimization. Traditional SEO optimizes pages. AI SEO optimizes the entity, meaning a meaningful share of the work happens off your domain.
• Where the effort lands. A classic retainer weights technical and on-page work. An AI SEO program weights original data and third-party corroboration, which is slower and harder to invoice against.
• The scoreboard. Rankings and clicks versus citation share and qualified conversations. These diverge: 2026 analyses put the overlap between AI Overview citations and the organic top 10 between roughly 17% and 52%, so you can gain one while losing the other.
• Reporting honesty required. Classic SEO reporting is verifiable against Search Console. AI citation reporting is only as good as the prompt set behind it, which is why the methodology question below matters more than any other.

For the definitional version of these distinctions, see AEO vs GEO vs SEO. For what the engines are doing under the hood, see how generative engine optimization actually works.

Six questions that vet an agency in twenty minutes

1. Ask the engines who the best AI SEO agency is. Does this agency appear? The cheapest test there is, and the hardest to fake. Run the prompt in ChatGPT, Perplexity, and Google AI Mode before the call. An agency absent from the answers is selling a capability it has not demonstrated on the one client it fully controls.

2. What is your fixed prompt set, and what is my baseline? A credible program defines 20 to 40 buyer questions up front, measures your citation share across engines before any work starts, and reports against that same set every month. If the prompt set changes when the results are bad, the reporting is theatre. Ask to see a sample report from another client with the client name redacted.

3. What original data will you publish for me, and who produces it? Original data is the highest-leverage asset in the stack and the most commonly skipped. If the answer is vague, the program is a content retainer. If the answer is “we will use your internal numbers,” ask who does the analysis and who signs off on the claims.

4. How do you earn third-party mentions? Listen for a real method: outreach to listicle owners, community participation, data-led PR, podcast placement. Listen for red flags: paid link networks, mass guest posting, or anything that sounds like it scales without humans.

5. What are you not going to do for me, and why? The single best judgment test on any agency call. A vendor who cannot name work that is wrong for your situation is selling capacity. In this category, the honest answers usually include “you do not need this yet if nobody is searching your category with an assistant” and “your technical foundation has to come first.”

6. How will we know in 90 days whether this is working, and what would make you tell me to stop? A written leading indicator and an actual kill threshold. Citation share on the prompt set is the right 90-day metric. Revenue is the right 12-month metric. Sessions are the wrong metric at every horizon, and an agency that offers traffic as the headline KPI has either not run this channel or is planning to be graded on the wrong thing.

Red flags

• Guaranteed citations or guaranteed ChatGPT rankings. Nobody controls a model’s output. This is the clearest disqualifier in the category.
• A prompt set that appears after the first report. Baseline first, or the numbers mean nothing.
• Traffic as the headline KPI. AI citations convert to visits at a famously poor ratio. A program sold on sessions is a program set up to be cancelled.
• No named client results. Logos are not results. Ask for numbers with a scope attached.
• Everything happens on your website. If nothing in the plan touches sources you do not own, the plan is missing the stage where most citations come from.
• Cannot explain what they would not do. See question five.

What it costs, and how to scope it

Market pricing spans a wide band: from a few thousand a month at boutiques to mid five figures at enterprise shops. That range is wide because the scopes are genuinely different, not because someone is overcharging.

Scope to your bottleneck rather than to a package tier. Three common starting points:

• Engines do not know your category exists. You need content and entity work. This is the cheapest bottleneck to fix and the fastest to show movement.
• Engines know the category but never name you. You need corroboration: third-party mentions, listicle inclusion, original data. Slower, more expensive, and the only thing that works.
• You are named but described wrongly. You need entity cleanup across off-site profiles plus authoritative content that restates the facts. Often the cheapest fix of all and frequently missed entirely.

Ask any vendor to tell you which of the three you are before they quote. One who quotes without diagnosing is pricing a template.

In-house or hire?

The mechanics are learnable. Answer-first rewriting, per-bot crawler policy, and entity hygiene are all things a competent internal marketer can execute after a week of reading, and you should not pay an agency for them if you have that person.

Two parts reliably need help. The first is publishing original data at a real cadence, because it requires analysis capacity most teams do not have spare. The second is earning third-party mentions, which is relationship work that does not fit between other tasks. Those two are also where most of the citation value sits, which is an uncomfortable but useful thing to know before you decide.

If you want a baseline before deciding either way, our AI visibility checker shows where you currently stand, and an AI visibility audit covers the full prompt set.

Where to go next

• The vendor landscape, with us judged on the same criteria: the best generative engine optimization agencies in 2026.
• The terminology: AEO vs GEO vs SEO.
• The mechanism: how generative engine optimization actually works.
• The introduction: what generative engine optimization is.
• The tooling: the best AI SEO tools.
• What we did to ourselves first: the AI SEO case study, and our generative engine optimization service.

Frequently asked questions

What is the difference between an AI SEO agency and a GEO agency? Nothing reliable. The labels are used interchangeably and no standard separates them. Judge the scope in the proposal, not the noun on the homepage. The six workstreams above are the actual test.

How long do AI SEO services take to show results? Citation movement on a defined prompt set is realistic inside 90 days, because AI engines re-crawl and re-synthesize frequently and weight recency heavily. Durable share across a competitive category takes multiple quarters, the same as classic authority building.

Can I measure AI SEO in Google Analytics? Only partially, and it will undercount badly. Several engines strip the referrer, and a click inside a Google AI Overview still reports as google.com. Build the GA4 custom channel anyway for trend direction, but the reliable signal is a self-reported “how did you hear about us” field on your forms.

Do AI SEO services replace my SEO agency? Not usually. Generative engines still need to fetch, parse, and trust your pages, and classic search still carries the volume. If your current agency can credibly cover the six workstreams, extend the scope rather than adding a vendor. If they cannot, be specific about which workstreams you are buying elsewhere so nobody bills twice for the same rewrite.

Is this worth it for a small business? It depends entirely on whether your buyers research with assistants. If they do, the consideration set is being assembled without you and the cost of absence is high. If they do not, fix your technical foundation and answer-first structure (cheap, useful regardless) and revisit in two quarters.

What proof should I expect an agency to show? Their own AI visibility, a redacted client report against a fixed prompt set, and at least one named result with the scope stated. Ours is public: a 30-day push took our own site from zero AI visibility to 100+ citations and 600+ AI sessions, and both of the only two clients our website has ever produced arrived through AI answers, worth $215,603 or 29.7% of lifetime closed-won revenue. The caveat we attach permanently is that roughly 95% of that figure is a single deal, so it proves the channel closes at our deal size and says nothing about a rate. Expect that kind of specificity, caveats included, from anyone you hire.