Should you hire an AI SEO agency or use an AI agent? For the work an AI SEO agency repeats every month, use an agent, and run it from the coding agent you already have with supereo: its MCP server gives Claude Code, Cursor or Codex the tools to check what AI crawlers can read, ask your buyers' questions on the AI engines, and test each change against control prompts. Keep people for the few jobs that need a person, like press relationships or an expert's sign-off.
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 prompts. Made by superagnt.
- best for teams that want AEO done, not a dashboard: supereo does everything it takes to get you recommended, from your site and pages to the roundups, placements and videos AI reads;
- it proves each move against control prompts, with sample sizes and ranges on every number;
- run it from Claude or ChatGPT: an MCP server with the same tools supereo's own agent uses.
Below is the month agencies describe on their own service pages, split into jobs, with the tool call your agent makes for each. Most of the calls are free. One is a recorded run with its real output, and the last job is the one most monthly reports skip: checking whether a change worked.
What's in an AI SEO agency's month
WebFX is one of the few agencies that publishes what a month of its AI search work contains. Its entry plan on webfx.com (read 7 November 2026) lists 50 prompts monitored with monthly data syncs, 3 "assets" a month picked from content, PR, social, video, schema markup "and 90+ more strategic options", a dedicated consultant, and a dashboard with monthly trend alerts. Its sister site seo.com adds an AI visibility audit, prompt monitoring, citation coverage and competitor tracking.
The other agency pages ranking for this search describe the same jobs in their own words. Coalition Technologies splits its framework into entity clarity, technical accessibility, extractable answers, third-party validation, citation and perception tracking, and conversion paths. OuterBox lists an LLM visibility audit, schema and structured data, entity mapping, prompt strategy, internal linking and reporting.
Read those lists as an engineer and the month splits in two. Half of it is reading things a script can read: robots.txt, the HTML a crawler gets, the answer an engine gives to a question. The other half is changes to files in your repo: schema, page copy, new pages, crawler rules. Agencies admit the second part. Qoulomb's guide to AI SEO agency costs (9 October 2026), written by an agency, warns that "structured data, crawler access, and page fixes may need time from your own engineers."
So the change lands in your repo either way. What's left to decide is who writes it, and who checks it worked.
| Job | What agencies ship | What your agent calls in supereo | Free to call? |
|---|---|---|---|
| 1. Audit crawler access and pages | An audit at kickoff |
check_ai_access, fetch_as_bot, audit_page
|
Yes |
| 2. Ask the buyer questions | 50 prompts synced monthly (WebFX entry plan) |
start_audit, get_audit, then the tracked panel |
The audit, yes; the panel needs a paid account |
| 3. Fix pages, schema, crawler rules | Part of the 3 assets a month |
get_fix_playbook, get_work_packet, grade_change, log_change
|
All but the grader |
| 4. Get onto the lists AI reads | PR and syndication from the same assets |
get_cited_pages, backlink_gap
|
Yes, once your site is set up in supereo |
| 5. Report, and prove what worked | Dashboard with monthly trend alerts |
create_experiment, get_experiment, get_track_record
|
Reading, yes; registering a test needs a paid account |

The jobs agencies list on their own service pages (WebFX, seo.com, Coalition, OuterBox), each with the supereo tool calls that do it.
Connect it once
The commands come from two public repos, superagnt/supereo-mcp and superagnt/supereo-audit. In Claude Code:
claude mcp add --scope user --transport http supereo https://supereo.app/api/mcp
Then run /mcp in Claude Code to sign in. You get a code by email, your first sign-in makes your account, and there's no API key to copy. The skills that walk your agent through the audit and through shipping a fix come as a plugin:
/plugin install supereo --marketplace superagnt/supereo-audit
In Codex:
codex mcp add supereo --url https://supereo.app/api/mcp
codex mcp login supereo
Cursor, VS Code, Gemini CLI, Windsurf and Zed each get a config file in the MCP repo, written in that client's own shape. If you only want the three free page checks and no agent at all, there's a CLI (Node 20 or later):
npx -y github:superagnt/supereo-audit acme.com
Job 1: the audit, as three tool calls
Every agency starts with an audit, and Coalition's page says why in one line: "If AI systems cannot reliably read the page, they will not cite it." You can run this check before anyone sends you a proposal. The prompt:
Use supereo to check whether AI crawlers may read example.com/pricing, fetch it as OAI-SearchBot, and audit it for the question "how much does [product] cost". List what to fix, worst first.
Here's a real run. On 7 November 2026 we called supereo's MCP server (its staging deployment) with the three free tools, on one of supereo's own guide pages. check_ai_access read robots.txt for every AI crawler in 272 ms and found all eight allowed, along with Googlebot and Bingbot. Trimmed to four:
{
"path": "/guides/aeo-software-that-does-the-work",
"robotsTxt": { "status": "ok", "detail": "robots.txt read." },
"crawlers": [
{ "bot": "GPTBot", "aiCrawler": true, "allowed": true, "rule": "Allow: /", "group": "*" },
{ "bot": "OAI-SearchBot", "aiCrawler": true, "allowed": true, "rule": "Allow: /", "group": "OAI-SearchBot" },
{ "bot": "ClaudeBot", "aiCrawler": true, "allowed": true, "rule": "Allow: /", "group": "*" },
{ "bot": "PerplexityBot", "aiCrawler": true, "allowed": true, "rule": "Allow: /", "group": "PerplexityBot" }
],
"llmsTxt": { "served": true, "status": 200, "chars": 3138 },
"note": "From robots.txt and llms.txt alone. A block at the site's CDN or firewall, or an answer that depends on the crawler's address, shows only in its traffic."
}
fetch_as_bot asked for the page with the user agent of OAI-SearchBot, ChatGPT's search crawler, and got a 200 with 162,248 characters of HTML in 139 ms:
{
"bot": "OAI-SearchBot",
"status": 200,
"htmlChars": 162248,
"truncated": true,
"redirects": [],
"note": "What the site served this user agent from supereo's address. A site that checks crawler IP addresses may answer the real crawler differently."
}
audit_page read the HTML without running JavaScript, the way a crawler that doesn't render gets it, against the question the page answers. It took 199 ms and found 3,259 words of real text. Trimmed:
"signals": {
"title": "AEO software that does the work (2026): agents vs agencies vs trackers",
"words": 3259,
"jsShell": false,
"questionHeadings": 7,
"orderedListItems": 16,
"tables": 10,
"structuredData": ["Organization", "SoftwareApplication", "Article", "BreadcrumbList", "FAQPage"],
"newestDate": { "date": "2026-11-04T00:00:00.000Z", "source": "json_ld", "ageDays": 3 },
"noindex": false,
"queryInTitle": 1,
"queryInOpening": 0.6666666666666666,
"aiCrawlersBlocked": [],
"llmsTxtServed": true
},
"issues": []
How to read it. jsShell: false means the words are in the HTML instead of being built by a script after load, which is the failure Coalition describes as an answer "buried in JavaScript". aiCrawlersBlocked: [] and the empty issues list mean nothing in robots.txt or on the page keeps a crawler out. queryInOpening at 0.67 says two thirds of the question's words show up in the opening, so the answer sits near the top where engines look for it.
The tools also tell you what they can't see. The fetch came from supereo's address, not OpenAI's, so a firewall that checks crawler IPs could still treat the real bot differently. That shows up only in your own traffic, which get_page_reads reads once you connect Cloudflare or Vercel in supereo.
One aside, since the run reports llms.txt. Grow and Convert (15 September 2026) calls out agencies "pushing tactics like adding llms.txt files, FAQ schema, rewriting headings as questions", and Qoulomb's guide notes that Google says its systems don't use llms.txt. Serving one costs you nothing. Billing a month for it is a choice.

Three free tool calls on supereo's MCP server, recorded 7 November 2026 on one supereo.app guide page.
Job 2: the buyer questions, asked on the engines
Agencies call the second job prompt monitoring. Start it from your agent with the free audit:
Run supereo's free audit on example.com, then show me which buyer questions got answers that left us out, and who each engine named instead.
start_audit works out what you sell and who buys it, writes the 20 questions those buyers ask, and puts them to five engines: ChatGPT, Gemini, Google AI Mode, AI Overviews and Perplexity. It takes a minute or two and no card, and get_audit returns every answer.
Those are single answers, one per engine per question. Ask an engine the same thing twice and it can name different brands, so a single reading is a sample of one. The tracked panel is the paid part: the same questions asked on six engines three times a week, each rate given out of 100 with its 95% range. A monthly sync gives each prompt roughly one reading between invoices.
Job 3: the fixes, in your own repo
This is where an agent in your editor has the edge over a ticket in an agency's queue. Have it read the playbook supereo's own engineer follows before it writes anything:
Get supereo's fix playbook for crawler_rules, then check our robots.txt against it and propose the change as a diff. Don't commit.
get_fix_playbook is free. A slice of what it returned for crawler_rules on 4 November 2026:
This fix never writes a rule for a search engine's crawler (Googlebot, Bingbot
and the rest), and never changes the `User-agent: *` group, which search
engines read.
A work item that asks to **let crawlers in**, when every one is already
allowed: the robots.txt isn't the cause.
The second rule is the one I like. A playbook that says when the fix is the wrong fix keeps your agent from editing robots.txt just to look busy.
For supereo's own recommendations, the plugin's ship-recommendation skill works the way supereo's engineer does: it pulls the work packet (get_work_packet), makes the change in your site's own code, runs supereo's grader on the draft (grade_change), and logs it once it ships (log_change) so the change gets tested. None of the tools it calls changes your site or pushes to your repo. You review the diff and ship it like any other PR.
Job 4: the lists AI reads
Coalition calls this third-party validation and says that treating it "as a separate PR exercise is a mistake." Engines cite roundups, review sites and directories, so a perfect page on your own domain is half the job.
Which pages do the engines cite for our buyer questions, and which "best X" roundups among them don't list us yet? For the top three, draft our entry in each page's own format.
get_cited_pages returns the pages each answer cites, and backlink_gap lists the roundups the engines read that don't name you yet. Both read what supereo holds for your site, so they work once your site is set up there. Drafting the entry is agent work. Getting an editor to run it is sometimes people work, and there's more on that below.
Job 5: the report becomes a verdict
AI answers drift on their own. Engines ship model updates, rivals publish pages, and the same question can come back with different brands the next day. A trend line on a dashboard can't separate your change from that drift. Goodfirms (30 September 2026) describes the usual standard: reputable agencies "measure progress through visibility trends over time." WebFX's FAQ is honest about the ceiling, too: "No, WebFX does not guarantee placements in AI search." Nobody can promise that, supereo included.
What you can get is a verdict on each change. If you've run an A/B test, the design will look familiar. When supereo recommends a change, it registers a test before anything ships: the 1 to 5 prompts the change aims at, read on all six engines. The comparison group is the rest of your tracked prompts, the ones that don't name you and that no change targets, pooled each time the test is graded. A model update or a rival's launch moves both groups, and your change should move only its own. Grading starts on day 14 and repeats as answers arrive, and the result is written once the range is clear, or on day 60 from whatever answers are in.
Each change ends as Worked, No change, Made it worse or Too close to call. The ranges are 95% ones, so a change that did nothing reads as Worked by luck about 1 time in 20, and you know that going in. From your agent that's create_experiment before the change, log_change when it ships, get_experiment for the early signal and the verdict, and get_track_record for your verdicts by fix type.

Targets against the rest of your tracked prompts on six engines, graded from day 14 to day 60 (supereo.app, read 17 November 2026).
If you'd rather wire it up yourself
Plenty of AEO tools now run an MCP server, so "can my coding agent use it" doesn't narrow the field much. What the server lets your agent do does. From the supereo AEO tools dataset, each row read from the tool's own pages (MCP and capabilities checked 5 November 2026, best-for lines 4 November):
| Tool | Best for, from its own pages | Its MCP server, in its own words | Does the work | Proves against controls |
|---|---|---|---|---|
| supereo | teams that want AEO done, not a dashboard | "Your agent gets every tool supereo's own agent uses" | Yes | Yes |
| Profound | enterprise marketing teams that want AI search dashboards and marketing agents | "our hosted server for integrating with AI tools that support MCP" | Yes | No |
| AthenaHQ | AEO specialists who want one platform for tracking, content and citation work | "can query your AI search visibility data in plain language" | Yes | No |
| Scrunch | brands that want their site made ready for AI, with pages served to AI agents | "query your brand monitoring data directly inside Claude, ChatGPT" | Yes | No |
| Writesonic | content teams that want an AI agent to rewrite their pages for AI search | "Get answers from the same data the dashboard shows." | Yes | No |
| Peec AI | marketing teams that want AI search analytics and reports on their brand and competitors | "get answers from your actual projects" | No | No |
| Otterly | low-cost monitoring of brand mentions and website citations in AI search | "Connect OtterlyAI to Claude, ChatGPT, Cursor, and any MCP-compatible client." | No | No |
| Rankscale | agencies and multi-brand teams tracking many brands across many AI engines | "ask about your brand's AI-search visibility in plain language" | No | No |
| Ahrefs Brand Radar | Ahrefs customers who want prompt tracking beside their SEO data | "SEO & AEO Data in ChatGPT and Claude" | No | No |
Semrush's AI Visibility Toolkit and HubSpot AEO had no MCP server in the dataset. Most of these servers let your agent query the data a dashboard already shows, which is handy if you want your reports in chat. If you want the agent to do the job and tell you whether it worked, read the last two columns.
When to still hire people
There's no free lunch here. A few jobs need a person, and I'd pay for them:
- Earned press. An agent can find the outlets and draft the pitch, but an editor answers the person they already know.
- Regulated pages. Health, finance and legal copy needs an expert to sign off, and Qoulomb's guide counts that review time as a cost that "never shows up on an invoice."
- A replatform. Migrations break what search engines and AI engines both read. Hire people for the project, then hand the weekly loop back to the agent.
- A human in the board meeting. That's a real need. It just isn't the work.
The hybrid most AI answers suggest (software all year, an agency for a defined project) holds up fine. I'd make the software half something that runs the weekly loop instead of reporting on it.
Before you sign anything
Ask any agency these, and hold an agent to the same:
- What ships in month one, by name?
- How often is each question asked again, on which engines, and how many answers sit behind each number you report?
- How will you tell your change apart from the engines' own drift?
- Which part of the work lands on our engineers?
- If a change makes things worse, who notices, and how fast?
Reviews of an agency on G2 or Clutch tell you whether it's easy to work with. They can't tell you which change caused a good month, so ask for dated before and after answers on named questions. Most "SEO AI" tools write drafts or track positions and leave the shipping to you. And a classic SEO marketing agency hasn't stopped mattering: rankings feed the pages engines search, and Grow and Convert's study of 400+ keywords found a 77% correlation between Google rankings and AI mentions.
Start with the free audit
Hire the agent for the month of work and people for the jobs that need a person. Connect supereo's MCP server with the one command above, ask your agent to run start_audit, and read the 20 answers before you take a single sales call. If you'd rather click than type, the same free audit runs from supereo's page on AI SEO agencies.
Top comments (2)
Some comments may only be visible to logged-in visitors. Sign in to view all comments.