You can automatically optimize WordPress content for SEO with AI by running a workflow that analyzes a post and rewrites its title, meta description, headings, and keyword coverage. AI Workflow Automation, a free WordPress plugin, includes an SEO optimization node that does this on publish or on demand, so every post ships search-ready without leaving your WordPress dashboard.
I wrote that node, and I want to be precise about what it does. It is a rewriting pass over text against a keyword list. It is not a ranking engine, and anyone selling you one is selling you something else.
AI Workflow Automation is a free WordPress plugin with a visual drag-and-drop builder that runs AI workflows, agents, and chatbots inside your own site, using your own API keys, with no external subscription required.
What can AI actually optimize for SEO, and what can it not?
AI can optimize the parts of SEO that are text transformation problems: phrasing a title within a sensible length, writing a meta description that reads like a human wrote it, restructuring headings so each one answers a real question, and checking that a target term and its variants actually appear in the body.
AI cannot optimize the parts of SEO that are not text problems. It cannot earn links, it cannot fix Core Web Vitals, it cannot make your site an authority on a subject you have no experience in, and it cannot decide which keyword is worth the effort. Those decisions come first and the model works inside them.
There is one more limit worth stating plainly. Google’s own title link documentation says title generation is “completely automated” on Google’s side, and it may rewrite your title using headings, visually prominent text, structured data, and anchor text. So the title you optimize is an input, not an output. Same with meta descriptions: Google’s snippet guidance says it “sometimes uses the meta description HTML element if it might give users a more accurate description of the page.” Sometimes. Not always.
Which SEO signals are worth automating?
Automate the signals that are repetitive, well defined, and cheap to check. Keep the ones that require judgment about your business.
| Signal | What the AI does | Automate it? |
|---|---|---|
| Title tag | Rewrites for clarity, front-loads the primary term, keeps it concise | Yes, with review. Google may still rewrite it. |
| Meta description | Writes a distinct, descriptive summary per page | Yes. Google explicitly says programmatic generation is acceptable if descriptions stay human-readable and diverse. |
| Heading structure | Reshapes H2s and H3s into question-form, answer-first sections | Yes. This is the highest-value automated pass. |
| Keyword coverage | Works the target term and its variants into the body naturally | Yes, with a hard cap. Google’s spam policies treat keyword repetition as a spam signal. |
| Excerpt and summary | Generates a short standalone summary of the piece | Yes. Low risk, consistently useful. |
| Internal links | Suggests relevant existing posts to link to | Partly. Have AI suggest, have a human confirm the target is genuinely relevant. |
| Keyword selection | Nothing. This is a business decision. | No. |
| Schema markup | Can draft JSON-LD, cannot validate it | Partly. Always validate the output before shipping. |
How do you build the AI SEO workflow in WordPress?
The build below runs on publish and takes about fifteen minutes. Open AI Workflows in your WordPress admin sidebar and create a workflow.
- Add a Trigger node and choose the WordPress core trigger, then select the
publish_postevent. The workflow now fires whenever a post is published. For an on-demand version, use a manual trigger instead and paste in the content you want optimized. - Add an AI Optimize for SEO node. In the builder it is labelled AI Optimize for SEO and sits in the AI Actions group. It has two fields: Content, which takes the text to optimize and accepts variable tags from upstream nodes, and SEO keywords, a comma-separated list of target terms.
- Wire the post body into Content. Reference the trigger output with a variable tag rather than pasting static text, so the node receives the actual published post on every run.
- Set the keywords. One primary term plus two or three genuine variants. Ten keywords in this field does not produce a post that ranks for ten keywords, it produces a post that reads like it was written for a crawler.
- Add an AI Model node for the metadata. The Optimize SEO node returns optimized body text, so if you want a separate title and meta description, generate them in their own node with an explicit instruction such as “return a title under 60 characters and a meta description under 155 characters, as two labelled lines.”
- Add a Human Input node set to Approval Required if the workflow will write anything back to a live post. I run this on every client site.
- Add a Create Post/Product node to write the result back. Set Post type to
post, choose your Post status, and use Field mappings to mappost_title,post_content, andpost_excerptfrom the correct upstream outputs. Each field needs its own source; do not wire one AI text block into three fields.
The Optimize SEO node costs 2 credits per run in cloud mode. In local mode it runs on your own OpenAI, Anthropic, or OpenRouter key and costs you only what the provider charges. The Create Post/Product node is free in both modes.
Should you optimize on publish, or in bulk?
Optimize on publish for new content and in bulk for the archive, because the two jobs have different failure modes.
On-publish optimization is cheap and safe. One post, one run, an editor sees the result. Wire it to the publish_post trigger and forget about it.
Bulk optimization of an existing archive is where people get hurt. Rewriting 400 published posts in one afternoon changes 400 URLs’ content simultaneously, which makes the result impossible to attribute: if traffic moves, you cannot tell which change caused it. It also risks rewriting pages that are already ranking well, which is the most expensive mistake in this whole category.
What I do instead: pull the twenty pages with high impressions and low click-through from Search Console, run those through the workflow, hold everything else, and wait four weeks. That is a readable experiment. A full-archive rewrite is not.
How does this work alongside Rank Math or Yoast?
It works alongside them, not instead of them. Rank Math and Yoast render the title tag, meta description, canonical, robots directives, and schema. AI Workflow Automation produces the text those plugins render. They are different layers.
The practical limitation, stated honestly: the Create Post/Product node maps to post_title, post_content, post_excerpt, registered post meta keys, and ACF fields. SEO plugin fields appear in the field mapping list only when they are exposed as registered post meta. If your SEO plugin’s meta description field does not show up, the workflow cannot write to it directly, and you have two options: generate the meta description, put it in front of an editor through a Human Input task, and paste it into the SEO plugin, or write it into a custom field your theme outputs.
I would rather tell you that than have you build a workflow that silently writes nothing.
How do you know whether the AI SEO pass actually worked?
Measure click-through rate first, position second, and never measure with a rewritten archive as your only variable.
Title and meta rewrites change how a result looks in the SERP, so their first-order effect is CTR at unchanged position. Open Search Console, filter to the pages you touched, compare the four weeks after against the four weeks before, and look at CTR before you look at anything else. If position moved too, that is a body-content effect and it takes longer to appear, typically one to three months.
Two controls make the reading trustworthy. Keep a holdout set of comparable pages you did not touch, so a seasonal swing does not get credited to your workflow. And annotate the date you ran the batch, because in six months you will not remember.
How do you optimize for AI Overviews, not just blue links?
Optimize for AI Overviews by making individual passages extractable, because AI answers lift passages, not pages. The unit of retrieval is a section, so each section has to make sense lifted out of the page with no surrounding context.
The numbers are why this matters. Seer Interactive’s tracking, covering 100+ client accounts and 7,800+ queries, found organic CTR of 2.94% on queries with no AI Overview against 0.84% when an AI Overview was present. Within AI Overview results, being cited as a source was worth 1.08% CTR against 0.6% when not cited (Seer Interactive). Seer’s later tracking through early 2026 shows AI Overview CTR recovering somewhat, so treat the exact figures as a moving target. The direction has been consistent: presence inside the answer beats presence below it.
Four structural changes do most of the work, and all four can go into your workflow’s prompt:
- Answer-first sections. The first sentence under each H2 answers the heading. No preamble.
- Question-form headings. “How much does X cost?” rather than “Pricing.”
- Named entities in every section. Never let “it” or “this” carry meaning across a section boundary, because the section gets extracted alone and the pronoun loses its referent.
- Facts with numbers in them. Declarative sentences containing dates, versions, and quantities get lifted. Marketing adjectives do not.
Add those four rules to the instructions in your Write an Article node and they apply to every draft the workflow produces, which is the entire point of automating this instead of remembering it manually each time.
Frequently asked questions
Can AI write meta descriptions that Google will actually use?
Sometimes. Google’s snippet documentation says it “sometimes uses the meta description HTML element if it might give users a more accurate description of the page than content taken directly from the page,” and it explicitly permits programmatically generated descriptions on large sites as long as they stay human-readable and diverse. Write one distinct, genuinely descriptive summary per page and avoid keyword strings, and it will be used more often than not.
Will AI-optimized content get penalized by Google?
Not for being AI-assisted. Google’s spam policies target scaled content abuse, which is generating many pages primarily to manipulate rankings rather than help users, and they apply regardless of whether content came from automation, humans, or both. An optimization pass over content you wrote for a real audience is not that. Mass-generating thousands of keyword variants of the same page is.
Does AI SEO optimization replace Rank Math or Yoast?
No. Rank Math and Yoast handle output: they render title tags, meta descriptions, canonicals, robots directives, XML sitemaps, and schema. AI Workflow Automation handles input: it generates the text those plugins publish. Run both. The AI workflow writes the copy, the SEO plugin puts it in the head of the document.
How many keywords should I put in the SEO keywords field?
One primary keyword plus two or three genuine variants. The AI Optimize for SEO node accepts a comma-separated list, and adding ten terms does not make a page rank for ten terms. It makes the model distribute mentions thinly across all of them, which produces text that reads like it was written for a crawler and covers none of them convincingly.
Try it on one post first
Install AI Workflow Automation from the WordPress plugin directory, drop in a manual trigger and an AI Optimize for SEO node, paste one underperforming post into the Content field, and read what comes back. If it is better, wire it to your publish_post trigger. If it is not, your keywords or your instructions need work, and that is a five-minute fix rather than a 400-post cleanup.
More worked publishing workflows are in the Content Creation archive.