I tested an agent-run SEO workflow on 10 WordPress sites over 90 days. Published pages went from 3 a month to 12, and the time I spent went from 6 hours a month to 1.5. The agent researched, wrote and published. What it could not do was show me, in one place, everything it had changed.
That gap is most of what this post is about. Agentic SEO is real and it works. The hard part is not getting an agent to do SEO. It is knowing what it did, approving the parts that matter, and putting things back when it gets one wrong.
- Agentic SEO is SEO work an AI agent does toward a goal, in a loop: read, change, check.
- The loop runs from goal to proposal to approval to the log, and undo closes it.
- Keep topics, drafts and anything visitors see with a person. Hand over the repeated typing.
- In WordPress: a limited agent login, approval for risky changes, undo after the chat is gone.
- Start with one repeated job, run it as a preview, and check the log the next day.
What is agentic SEO?
Agentic SEO is SEO work done by an AI agent that is given a goal, not a single prompt: it reads your site, decides what to change, makes the change and checks the result, in a loop. You set the goal and the limits, and you approve what matters.
Search Atlas defines it as “a search optimization model where autonomous AI agents plan, execute, and iterate SEO actions”. SEO.com puts it more simply: “AI agents plan and execute SEO tasks on their own.” The word that does the work is “execute”.
Agentic SEO vs asking ChatGPT for a meta description
A chatbot writes a meta description for you to paste.
An agent finds the 40 posts with no meta description, writes them and saves them.
The other “agentic SEO”: optimizing for AI agents
You will also see the term used the other way round: making your site easy for AI agents to read and cite. First Page Sage and Addy Osmani’s agentic-seo tool use it that way. This post is about the first meaning: agents doing the SEO.
How agentic SEO differs from traditional SEO
| Traditional SEO | Agentic SEO | |
|---|---|---|
| Who does the work | A person, field by field | An agent, in a loop |
| Cadence | Audits and projects | Ongoing, job by job |
| Where changes are made | By hand in wp-admin | Through the SEO plugin, by the agent |
| Who checks | The same person | You, before risky changes land |
| How mistakes get fixed | Find and retype | Undo from the change log |
How an AI agent SEO workflow runs
Seven steps, and the loop starts again. One of them stays with you.
- You set a goal and limits.
- The agent reads the site: pages, SEO fields, site check issues.
- It proposes changes, with the old value next to the new one.
- Checks run on each change.
- You approve what waits for you.
- The change is applied and logged.
- You measure, and undo anything that went wrong.
Then the loop runs again on the next job.
What agents can do for SEO today
- Missing meta descriptionsFinds every post without one and writes it.
- Long titlesRewrites titles over 60 characters and keeps the keyword.
- FAQ schemaAdds FAQ markup to help articles and fixes what the check flags.
- Alt textFills empty alt text on images.
- Broken linksFinds them and unlinks or fixes them.
- DraftsWrites a new post and saves it as a draft for you to read.
Agentic SEO in WordPress: the setup that keeps you in charge
In WordPress, the agent needs a way in. Most setups give it an application password on an admin account and hope. The official WordPress guidance on the MCP Adapter says the opposite: “Especially in production, create a specific role/user with limited capabilities.”
Give the agent its own login, not yours
- Edit titles, descriptions, schema, redirects and alt text
- Write drafts and edit posts (edits to live posts wait for you)
- Run site checks and read the results
- Touch users or roles
- Install plugins or change themes
- Edit header and footer code
Preview first, then approve by area
Any job can run as a preview first, the old value next to the new one. You pick which areas wait for your approval.
The changes that should always wait
Search Atlas sorts agent actions into three risk tiers, with high-risk work that “requires full approval before deployment”. In WordPress that maps to settings you can see: robots.txt edits, noindex and trash always wait, whatever the agent asks.
Undo after the chat is gone
Every change lands in a log with who made it and when, so you can undo one change, or a whole batch, a week later.
A worked example: one job, start to finish
Write meta descriptions for every post that doesn’t have one. Show me the preview first.
Risks of agentic SEO, and the fix for each
| Risk | What goes wrong | The guard |
|---|---|---|
| Wrong change at scale | One bad rule hits 200 posts | Preview first, batch undo |
| Instructions hidden in a page | The agent follows text it read on your site | Server instructions tell agents to ignore them; you still review |
| Too much access | The agent can install plugins or add users | A limited SEO role, never admin |
| Thin AI content | Drafts go live unread | Drafts wait for you |
| Changes nobody can trace | Settings “changed on their own” | A log with who, when and how |
What I learned running it on my own sites
I ran the test on 10 of my own WordPress sites for 90 days, before AgentedSEO existed. It was my test, not a customer result.
The agent did the repeated work well. It researched keywords from Search Console, wrote the drafts and published them. Published pages went from 3 a month to 12, and my time went from 6 hours a month to 1.5.
Two things still needed me. Schema was the one job the agent needed another plugin for. And I could not see at a glance what it had changed.
Those two gaps are what AgentedSEO is built around: schema that is checked before it saves, and every change shown first, logged with a name and a time, and one click from undone.
How to start with agentic SEO this week
- 1Pick one repeated job (missing meta is a good first one).
- 2Connect your agent with its own login: Claude, Cursor, Codex, Hermes.
- 3Run it as a preview, approve, and check the log the next day.
FAQ
SEO done by an AI agent that works toward a goal: it reads your site, makes changes and checks them, instead of answering one prompt at a time. You set the limits and approve what matters.
In traditional SEO a person finds the issue and makes the fix, field by field. In agentic SEO the agent does both, in a loop, and the person sets goals, approves risky changes and checks results.
No. It replaces the repeated typing: meta on every post, schema, long titles, alt text. Choosing topics, judging drafts and deciding what is worth ranking for stay with people.
Agentic AI is the general idea of AI that acts toward a goal with tools. Agentic SEO is that idea applied to search work, with tools that can read and change a site’s SEO.
Not by itself. It gets more of the basics done, more often. In my own 10-site test the clear gain was output: 3 to 12 published pages a month.
It can be, if the agent has its own limited login, risky changes wait for you, and every change can be undone after the chat ends.
The agent you already pay for plus an SEO plugin it can operate. Several SEO plugins include agent access in their free versions.
Try agentic SEO on a site you can’t break.
Get a WordPress sandbox with AgentedSEO installed, connect your agent, run one job, and undo it.
Early Adopter Lifetime until 1 December 2026, 23:59 PT · 15-day refund (80% back)