AI SEO agency or AI agent? An agency's month, task by task, with the prompts that run it

Should you hire an AI SEO agency or use an AI agent? Use an agent for the recurring work an AI SEO agency bills you for each month (asking your buyers' questions in AI engines, fixing the pages AI reads, building the pages it's missing, getting you onto the lists it cites), and pay people only for the parts that need a person in the room.

That's my answer, and I'm biased: I build supereo, which is one of those agents. So this post shows the work instead of arguing it. Below is an AI SEO agency's month, task by task, taken from one agency's published plan, next to the prompt an agent runs for each task from Claude or ChatGPT, a real recorded run, and the check most agency reports skip.

We put this exact question to ChatGPT, Gemini, Perplexity, Claude and Google's AI Mode on 12 November 2026. All five drew the same line: software tells you where you stand, an agency does the work. This post is about the space between those two.

An AI SEO agency's month, task by task

Most agencies don't publish what a month contains. WebFX does, which makes it a useful baseline. Its entry AI search plan is $3,000 a month on a 6-month commitment (webfx.com, read 7 November 2026). For that you get 50 prompts monitored with data synced once a month, 3 "assets" a month picked from a menu of content, PR, social, video, syndication, schema markup and 90+ other options, a dedicated consultant, and a dashboard with monthly trend alerts. Its AI SEO services page on seo.com (read 12 November 2026) adds the visibility audit, prompt monitoring, citation coverage and competitor tracking.

Every other agency page I read for this (Coalition, OuterBox, SEO.co, Qoulomb's cost guide) lists roughly the same six jobs. Here they are, with what an agent does instead.

The agent here is supereo. supereo (supereo.app) is an AI agent that does everything it takes to get your brand recommended by ChatGPT, Gemini, Perplexity, Claude and Google, on your site and off it: site fixes, new pages, roundups, placements and video. It proves what worked against control questions. Made by superagnt. Its MCP server exposes the same tools its own agent uses, so you can drive every row above from Claude or ChatGPT.

The prompts, one per task

These are the prompts I'd type into Claude Code or ChatGPT with supereo's MCP server connected. Swap in your own domain, phrases and rivals.

Task 1, the questions:

Use supereo's keyword_data to pull US monthly searches for these phrases, then turn the ones with buying intent into 20 questions a buyer would ask ChatGPT about my category: [your phrases]

Task 2, the audit:

Check whether AI crawlers can read example.com/pricing, fetch it as OAI-SearchBot and Claude-SearchBot, and audit it against the question "how much does [product] cost". Tell me what to fix first.

Task 3, a fix:

Diagnose example.com/compare/us-vs-rival against the pages AI engines cite, then read the fix playbook for the top gap and write the change.

Task 4, a missing page:

Which of my 20 buyer questions has no page on example.com that answers it in its first two sentences? Draft that page, with a comparison table and a source for every number.

Task 5, off site:

Run backlink_gap for [your category]: which "best X" roundups do the engines read that don't list us yet? For the top five, draft our entry in each page's own format.

Task 6 is the one you don't prompt. It's further down.

A real run: the audit step from Claude Code

Prompts are cheap, so here's one that actually ran. On 4 November 2026 we started Claude Code 2.1.299 on claude-opus-5-5 with no built-in tools and supereo's MCP server (staging) as its only one, and typed:

Use supereo to check whether AI crawlers can read supereo.app, then look at https://supereo.app/about the way ChatGPT's search crawler sees it, and tell me what to fix first so AI answers cite it. Keep the answer short.

Claude picked three calls on its own, each on the /about page:

check_ai_access                                   360 ms
fetch_as_bot     bot=OAI-SearchBot                190 ms
audit_page       query="What is supereo?"         239 ms

The first came back with every crawler allowed. A slice of it:

{ "bot": "OAI-SearchBot", "aiCrawler": true, "allowed": true, "rule": "Allow: /", "group": "OAI-SearchBot" },
{ "bot": "Claude-SearchBot", "aiCrawler": true, "allowed": true, "rule": "Allow: /", "group": "Claude-SearchBot" },
{ "bot": "PerplexityBot", "aiCrawler": true, "allowed": true, "rule": "Allow: /", "group": "PerplexityBot" }

The fetch got a 200, server-rendered. The audit read the page the way a crawler without JavaScript gets it (trimmed):

"signals": {
  "h1": "About supereo",
  "words": 480,
  "jsShell": false,
  "questionHeadings": 5,
  "structuredData": ["Organization", "Person", "SoftwareApplication", "Offer", "UnitPriceSpecification", "AboutPage", "WebSite"],
  "queryInTitle": 1,
  "queryInOpening": 1,
  "aiCrawlersBlocked": [],
  "llmsTxtServed": true
},
"issues": []

Claude's reply led with the thing a kickoff deck usually buries on slide 30:

Fix first: the brand's own "sameAs": [] is empty in both the Organization and SoftwareApplication data. Only the founder has profiles linked.

Then it pointed off site: get onto the roundups the engines read (that's backlink_gap), and connect the site's Vercel traffic so real crawler reads show up, since a fetch from supereo's own address can't rule out a block at the edge.

Total tool time was 789 ms. That's the audit an agency books a kickoff call for. It's also the cheap part. The expensive part is knowing whether the fix worked.

The check an agency report can't make

AI answers move on their own. Engines ship model updates, rivals publish pages, and the same question asked twice can get two different answers. So a report saying you're up this month tells you where you stand. It doesn't tell you which change did it.

Two things fix that, and both come down to counting.

Read more often. WebFX's entry plan syncs its 50 prompts once a month: 12 readings per prompt per year. Weekly syncs are an add-on at $150 a month per 50 prompts. supereo asks every question on six engines three times a week on every plan: 156 readings per question per engine per year, 936 across the six. Twelve data points can't separate a change from noise.

Test against controls. Before supereo ships a change, it registers a test: 10 target questions the change should move and 10 control questions it shouldn't, asked on one engine. The controls soak up everything the change had no part in (a model update, a rival's launch, a quiet week). After 45 days, once every question has 25 runs, it computes the verdict once and records it, whatever it says. If the runs aren't all in by day 60, it records the verdict from what it has. Each change ends as Worked, No change, Made it worse or Too close to call.

If you write software, you know this shape: a holdout group, a pre-registered test, a fixed read date. It's the obvious way to measure a change. Somehow it's missing from almost every AI SEO report.

Nobody guarantees placement, and nobody should. WebFX says so in its own FAQ ("No, WebFX does not guarantee placements in AI search"), and so do SEO.co and Goodfirms. What you can get is a verdict per change instead of a trend line.

What the first year costs

WebFX's prices are from webfx.com (read 7 November 2026), and its published ladder climbs to $68,250 a month. Qoulomb's are from its own cost guide (dated 9 October 2026). supereo's are its billing today. For the wider market, Goodfirms puts GEO agency retainers at $1,500 to $50,000+ a month (Goodfirms, updated 30 September 2026).

Copilot writes each change for your team to ship. Autopilot ships them itself, on WordPress, Webflow, Shopify, Wix, Framer, Next.js or Astro sites. The trial takes a card up front and charges nothing until the week ends.

When to still hire people

There is no free lunch here. An agent is the right call for the weekly loop, and people still win at a few things:

• Journalist relationships. Earned coverage in a trade outlet runs on a person an editor already trusts. An agent can find the outlets and draft the pitch. It can't buy the editor a coffee.
• A replatform or a rebuild. Migrations break things Google and AI engines both read. Hire people for the project, then hand the weekly loop to the agent.
• Regulated content. Healthcare, finance and legal pages need an expert to sign off. Qoulomb's guide makes the same point about review time that never shows on an invoice.
• A person in the meeting. If your board wants a human presenting the AI search plan, that's a real need. It just isn't the work.

The hybrid most of the engines suggested (software all year, an agency for a defined project) is fine. I'd only swap the software from a tracker that reports where you stand to an agent that does the weekly work.

Questions to ask before you sign

If you're hiring an AI SEO agency anyway, ask these. They work on agents too.

1. How will you tell a change's effect apart from the engines' week-to-week drift?
2. How often do you re-ask each question, and on how many engines?
3. What sample sits behind each number you report?
4. What ships in the first month, by name?
5. If a change makes things worse, who notices, and how fast?
6. How long is the commitment, and what does leaving early cost?

FAQ

What do AI SEO services include? An audit of how AI engines answer your buyers' questions, prompt monitoring, schema and entity fixes, answer-first content, digital PR to get mentioned on the sites AI cites, and reporting. The table at the top maps each one to an agent task.

What's the best AI SEO agency? The one that ships work every week, publishes its price, and can tell you which change moved an answer. If nobody on your shortlist can answer question 1 above, an agent that tests against controls is the better buy.

Are AI SEO agency reviews worth reading? For service quality, yes. For results, read them with care: a review of a good month can't tell you which change caused it. Ask the agency for before and after answers on named questions, with dates.

Is there a list of AI SEO agencies? Plenty. Grow and Convert, Embarque, OneLittleWeb and Goodfirms all publish one, and most are written by agencies that put themselves first. Use them for a shortlist, then ask the six questions.

How much does an AI SEO agency cost in the USA? Published US prices start around $2,000 to $3,000 a month (Qoulomb, WebFX) and run past $50,000 a month for enterprise programs (Goodfirms). Expect a minimum term: WebFX's is six months.

How is an AI SEO agency different from an SEO marketing agency? An SEO agency works on Google rankings. An AI SEO agency adds the answers in ChatGPT, Gemini, Claude, Perplexity and Google's AI Overviews and AI Mode to the same retainer. Rankings still feed those answers, so good AI SEO keeps the SEO.

Is SEO AI software the same as an AI agent? No. Most SEO AI tools write drafts or track positions and leave the shipping to you. An agent runs the loop: asks the questions, makes the change, checks whether it worked.

How long until results? Agencies quote two to six months for the first citations (Qoulomb, Goodfirms). A supereo verdict takes 45 to 60 days per change, since a single good week is just noise.

Start where an agency starts

Every agency opens with an audit. supereo's is free and needs no card: it writes the 20 questions your buyers ask, puts them to five AI engines, and shows every answer and who gets picked instead of you. Run it from supereo's AI SEO agency page, then decide whether you need people, an agent, or both.

I build supereo, and I wrote this with AI help from our own research data.