Search for the best AI SEO agency and you will find a dozen ranked lists. Read them closely and a pattern appears. Almost every list was written by an agency, and almost every agency placed itself first.
One list scores its author 97 out of 100, then ranks four competitors at 94, 91, 88 and 85. No method is published. Nothing explains what those numbers measure. Another agency calls itself the number one AI SEO agency on its own website. A third says it is ranked first, on a page it wrote.
This article takes a different approach. We published our criteria before we looked, applied the same tests to every agency including our own, and recorded what we actually found. Where a test could not be run, we say so.
Note - Disclosure: Rizing Metrics is our own agency. We have included it in this comparison and applied exactly the same tests. You can verify every result yourself, and the section below explains how in about two minutes per site.
What makes an AI SEO agency worth hiring?
An agency worth hiring can name the queries it will track, show you where you stand on them today, and report the change on a schedule. Everything else is presentation. An agency that cannot describe its measurement method is selling you activity rather than outcomes, however confident the marketing sounds.
That principle produced the four tests below. Each one is objective, each one takes minutes to run, and none of them depends on trusting what an agency says about itself.
Why the usual signals fail here
Choosing a traditional SEO agency is hard enough. Choosing an AI SEO agency is harder, because the field is roughly two years old and almost nobody has multi-year results to point at. Case studies are thin everywhere, including here. Awards mean nothing yet. Client lists prove that somebody paid, not that the work succeeded.
Team size tells you little, since a three-person team with a clear method will outperform a thirty-person team applying templates. Price tells you less than people assume, because the cheapest offers are templated and the most expensive are often traditional SEO with new labels on the invoice.
That leaves evidence you can verify yourself, which is what this comparison is built on.
How we compared them
Four checks, applied to every agency on 28 September 2026.
| Test | What it shows | How it is checked |
|---|---|---|
| Structured data on their own site | Whether they implement the markup they sell | View page source, search for application/ld+json |
| AI crawler access in robots.txt | Whether AI systems are allowed to read them | Visit their domain followed by /robots.txt |
| An llms.txt file | Whether they publish a machine-readable summary | Visit their domain followed by /llms.txt |
| Published measurement method | Whether success is defined or implied | Read their service page |
These tests do not measure how good an agency is at its job. A brilliant practitioner can have a neglected website. What the tests do measure is whether an agency applies its own advice to itself, which is the only thing a prospective client can verify without becoming a client first.
How AI search actually picks which businesses to name
Understanding the mechanism makes the tests below make sense, and it takes two minutes to explain.
An AI system answering a question about businesses does two distinct things. First it retrieves candidate sources, usually through a search index. Bing supplies Copilot. Google supplies its own AI Overviews. Perplexity and ChatGPT run their own retrieval over a search layer. Second it synthesises what it retrieved into a single answer, choosing which businesses to name and how to describe them.
Both steps can go wrong for a business, and they fail differently.
| Step | What goes wrong | What fixes it |
|---|---|---|
| Retrieval | Your pages are never fetched as candidates | Crawler access, technical health, content that matches the question |
| Synthesis | Your pages are fetched but you are not named | Clear claims, structured data, corroboration from other sources |
Most agencies work only on the second step, because it looks like content marketing and content marketing is what they already sell. The first step is where blocked crawlers, missing structured data and inconsistent business information quietly remove a company from consideration before any content is ever judged.
That is why an agency blocking AI crawlers on its own website is worth noticing. It suggests the first step is not part of how they think.
What we found
Results as recorded on 28 September 2026. Any of these can change, and several probably will once this article is published.
| Agency | Structured data | FAQ markup | llms.txt | AI crawlers |
|---|---|---|---|---|
| Searchbloom | Extensive, including ItemList | Yes | Yes | Allowed |
| Percepture | Extensive, including ItemList | Yes | Yes | Allowed |
| Rizing Metrics | ProfessionalService, WebSite, FAQPage | Yes | Yes | 24 agents named and allowed |
| Victorious SEO | Organization, FAQPage | Yes | No | Allowed |
| ZipTie.ai | Article, Organization, Person | No | No | Allowed |
| OuterBox | WebPage, Organization, Place | No | Yes | Allowed |
| Onely | WebPage, Organization | No | Yes | Several blocked |
| Directive Consulting | WebPage, WebSite | No | Yes | Allowed |
Four further agencies returned a 403 response and could not be inspected. We list them separately below rather than scoring them on evidence we do not have.
The result that surprised us
Onely publishes a headline on its homepage reading “Be seen by customers in AI search”. Its robots.txt file blocks GPTBot, ClaudeBot, anthropic-ai and CCBot.
Context matters here, and the nuance is worth understanding because almost nobody explains it. AI companies run two different kinds of crawler. Training crawlers gather text to train future models. Search crawlers fetch pages in real time when a user asks a question. Blocking the first while allowing the second is a legitimate editorial position, and Onely does allow ChatGPT-User, which is a search-time agent.
Blocking ClaudeBot is more consequential, because that agent serves search as well as training. The wider point stands regardless: an agency selling AI search visibility has made deliberate choices about which AI systems may read it, and most visitors would never think to check.
Note - Run this check on any agency you are considering. Type their domain followed by /robots.txt into your browser. Look for GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot and Google-Extended. A Disallow under any of those names means that system is being told to stay out.
Best AI SEO agency overall
Searchbloom and Percepture score highest on evidence. Both implement extensive structured data including ItemList markup, both publish FAQ markup, both maintain an llms.txt file, and both leave AI crawlers unrestricted. Their own pages are built the way an AI-optimised page should be built.
Neither publishes a measurement method, which is the gap shared by every agency we checked. Ask them which queries they track and what your baseline is before you sign.
Look at what those two get right, because it is repeatable rather than mysterious. ItemList markup tells a machine that a page contains a ranked set and what is in it. FAQ markup pairs questions with answers in a form that can be lifted directly. An llms.txt file gives a model a plain summary of what the site is and where the important pages sit. None of that is exotic, and most agencies simply have not done it.
What neither publishes is a measurement method. No named queries, no baseline, no reporting schedule. That gap is universal across every agency we checked, including the ones that could not be inspected, and it is the single most useful thing to press on during a sales call.
Best AEO agency
Answer Engine Optimization depends on question-form content paired with FAQ markup, so the test is direct. Does the agency implement FAQPage schema on its own site?
Four do: Searchbloom, Percepture, Victorious SEO and Rizing Metrics. Four do not, including two that sell answer engine optimization as a named service. An agency selling answer optimization that has not marked up its own answers has not implemented what it is proposing to sell you.
Worth being precise about what FAQ markup does, since it is the most misunderstood item on this list. It does not make you rank. What it does is state, in a format that requires no interpretation, that a specific question has a specific answer on your page. An answer engine looking for something to quote finds a labelled pair rather than having to infer one from prose.
Visible content and markup must match. Marking up questions that do not appear on the page violates Google structured data guidelines and can produce a manual action. Any agency proposing to add FAQ schema without adding the visible questions is proposing something that can get you penalised.
Ask a prospective AEO agency two things. Which questions will you add, and where will they appear on the page? Vague answers here usually mean schema is being bolted on rather than content being written.
Best GEO agency
Generative Engine Optimization is about being cited inside AI-generated answers. Research published at ACM SIGKDD in 2024 tested nine content methods and found the three most effective were adding quotations from credible sources, adding statistics in place of qualitative claims, and citing sources, improving visibility by 41, 32 and 28 percent respectively.
Judged on whether their own content follows that pattern, ZipTie.ai and Searchbloom stand out. Both publish long, heavily evidenced articles that cite their sources. ZipTie carries no FAQ markup, which costs it on the AEO test, but its content is written the way the research says generative engines reward.
One finding from that same research deserves attention if you are not the biggest name in your market. Citing sources improved visibility by 115 percent for websites at position five, while reducing it by 30 percent for sources already ranked first. Evidence-led content helps challengers considerably more than it helps incumbents.
The practical takeaway from that research is unusually actionable. Content built on specific, attributed claims outperforms content built on confident assertion. An article saying that most businesses see results in three to six months, with no source, gives a model nothing to work with. An article citing a study, naming it, and stating what it measured gives a model something quotable.
This is also why thin content fails in AI search even when it ranks acceptably in traditional search. A three-hundred-word page can rank for a long-tail query. It rarely contains anything a generative engine wants to quote.
Best for local businesses
Local AI visibility depends on entity consistency rather than content volume. The business name, address and phone number must agree across every source a model might read, and the Google Business Profile must be complete and active.
Most agencies on this list are national or enterprise-focused and do not treat local entity work as a distinct discipline. Rizing Metrics does, which is our own bias and worth stating plainly. If you run a local service business, the question to ask any agency is how it handles inconsistent business information across directories, because that is the work that decides whether an AI tool names you.
Local AI visibility has one failure mode that dominates all others, and it is worth understanding before you hire anyone to fix it.
A model asked to recommend a plumber in a specific town assembles its answer from several sources at once. Your website, your Google Business Profile, directory listings, review platforms and local citations all contribute. When those sources disagree, even slightly, the model has a problem. Two different phone numbers, an old suite number, a business name that appears three ways, an address that moved two years ago and still appears in four directories.
Faced with contradiction a model does one of two things. It hedges, describing you vaguely, or it omits you and names a business whose details are consistent. Neither outcome is caused by weak content.
The work that fixes it is unglamorous. Find every place your business appears, record what each one says, correct the ones that disagree, and keep them aligned afterwards. Ask any local-focused agency how it does that, and whether it audits directories it does not control.
Best for B2B
B2B buyers use AI tools early in the process, while building a shortlist rather than while making a purchase. A company never named at that stage never enters the evaluation.
Directive Consulting is explicitly B2B positioned. Its own site carries minimal structured data, which is a real gap, though its B2B specialism is genuine and its content addresses buyer-committee dynamics that generalist agencies overlook.
B2B has a second characteristic that changes the work. Buying decisions involve several people, and they ask different questions. A technical evaluator asks about integration and implementation. A finance approver asks about cost and contract terms. An executive sponsor asks about risk and outcomes.
Each of those is a different query, and each can be answered on a different page. A B2B AI visibility programme that only targets the phrases a marketing lead would type covers one member of a committee that may have five.
The agencies we could not verify
Four agencies returned a 403 response to an automated request, which prevented inspection. They are Thrive Agency, Coalition Technologies, Commerce Pundit and Intero Digital.
Blocking automated traffic is a normal security measure and carries no implication about quality. It does mean we cannot confirm or refute their claims, and several of those claims are substantial. One describes itself as the number one AI SEO agency and cites more than 800 case studies. Ask to see them.
How to run this comparison yourself
Every test in this article takes about two minutes per agency and needs no tools beyond a browser.
- Open the agency homepage, right-click, choose View Page Source, then search the page for application/ld+json. No result means no structured data
- Search the same source for FAQPage. An agency selling answer optimization should have it
- Type their domain followed by /robots.txt and look for Disallow rules under GPTBot, ClaudeBot, PerplexityBot or OAI-SearchBot
- Type their domain followed by /llms.txt and see whether anything loads
- Ask three AI tools what the agency does and compare the answers against each other
That last check is the most revealing. An agency that cannot get AI tools to describe it consistently is unlikely to achieve that for you.
A worked example
Take any agency you are considering and spend two minutes on it before reading a word of their sales page.
Open the homepage and view the page source. Search for application/ld+json. If nothing appears, that agency publishes no structured data, which means it has not implemented the most basic machine-readability measure on the one website it fully controls. Search the same source for FAQPage. Absence is not fatal, but an agency selling answer optimization should have it.
Now type the domain followed by /robots.txt. Read the user-agent blocks. A line reading Disallow with a forward slash beneath GPTBot means OpenAI crawlers are told not to read that site. Then try the domain followed by /llms.txt. Most sites return a 404, and a file appearing there tells you the agency is at least tracking emerging standards.
Finally, open ChatGPT, Gemini and Perplexity and ask each one what the agency does. Compare the three answers. Inconsistent descriptions, confusion with a similarly named company, or a model that cannot describe them at all, all point at the same underlying problem, and it is the problem they are proposing to solve for you.
Questions to ask before you sign
Five questions separate practitioners from rebranded generalists.
- Which specific queries will you track, and how often? A real answer names queries and a schedule
- What is my baseline today? An agency that has not measured your starting point cannot demonstrate progress later
- What does your own AI visibility look like? Ask them to show you, not tell you
- What can you not promise? An honest answer names the limits of attribution and platform variability
- What happens when a platform changes how it selects sources? Strengthening fundamentals is the correct answer. A proprietary trick is not
Six claims should end the conversation:
- Guaranteed ChatGPT rankings or guaranteed AI recommendations
- Any offer to place your business inside an AI model training data
- Research statistics presented as their own client results
- A proprietary visibility score with no published calculation
- No named queries, no baseline, no reporting schedule
- No verifiable clients, case studies or reviews
What does the work cost and how is it priced?
Most agencies on this list publish no pricing, which makes comparison difficult and is itself worth noticing. Three structures are common, and the structure matters more than the figure.
| Structure | How it works | Suits |
|---|---|---|
| Monthly retainer | Continuous work, reported on a cycle | Most businesses, since this work is ongoing |
| One-off audit | Assessment and roadmap, implemented by you | Teams with in-house capacity |
| Audit then retainer | Paid assessment first, retainer if you continue | Testing an agency before a long commitment |
Four variables drive what you pay. How much of the foundation already exists, since a site with clean structured data and consistent business information needs far less remediation. How many locations and services require coverage. Whether content production is included or supplied by you. Whether ongoing measurement is included, which is labour-intensive when done properly and frequently omitted when it is not.
Offers priced far below the market are almost always templated, meaning the same page structure with your city dropped into it. Premium pricing is only justified by measurement, so an agency charging enterprise rates should be able to name the queries it tracks and state your baseline. An agency that cannot is charging for positioning rather than work.
Note - One question settles more than any pricing table. Ask what the first thirty days include. A credible answer describes an audit, a baseline measurement and a prioritised list of fixes. A vague answer means the engagement has not been planned.
What should you expect in the first ninety days?
Timelines in this field are quoted loosely, so it helps to know what is actually happening in each phase and what evidence should exist at the end of it.
| Period | Work | Evidence you should see |
|---|---|---|
| Weeks 1 to 4 | Audit, baseline measurement, technical and crawler fixes | A written baseline and a prioritised fix list |
| Weeks 5 to 8 | Structured data, entity corrections, first content | Corrected listings, markup live, pages published |
| Weeks 9 to 12 | Content depth, third-party corroboration | First measurable movement in mention rate |
A business starting with accurate information, reasonable authority and a healthy website can see change within the first weeks. A business starting with inconsistent listings, a thin backlink profile or blocked crawlers should expect several months, because the foundation has to exist before content can do anything.
Any agency promising visible AI results in the first fortnight is either describing a business that was already close, or describing something that will not happen.
What this comparison cannot tell you
Four tests on a public website do not measure competence. They measure whether an agency applies its own advice to itself, which is a useful signal and a limited one.
What we could not assess includes client results, retention, the quality of strategic thinking, and whether an agency is a good fit for a particular business. None of that is visible from outside, and any list claiming to rank it from public data is guessing.
We also cannot measure revenue attribution, and neither can anyone else. When an AI tool names a business, the customer usually searches the brand directly or simply calls. No referrer is passed and no click is recorded. Any agency claiming to prove that a specific sale came from a ChatGPT mention is describing attribution that does not exist.
How to use this list
Shortlist three agencies. Run the four checks on each one, which takes under ten minutes in total. Then ask all three the same five questions and compare the answers rather than the proposals.
The agency that names its queries, shows your baseline and commits to a reporting schedule is the one doing the work. Scores out of 100 tell you nothing that you cannot verify better yourself in two minutes.
Rizing Metrics is an AI SEO agency working on answer engine optimization, generative engine optimization and entity consistency as a single scope. We have included ourselves in this comparison and applied the same tests, and you are welcome to run them again. Our answer engine optimization service explains how the work breaks down, and AEO vs GEO vs LLM SEO agencies covers what the different labels actually mean.
