# Agentic SEO: what it is, and the exact run that produced this post > Agentic SEO explained through a real run: what the agent did with Search Console and keyword data, where a person stepped in, and what can go wrong. Source: https://contengi.com/blog/agentic-seo Author: Stefan Maritz Published: 2026-10-16 **TL;DR** - Agentic SEO means handing an AI agent an SEO goal plus the tools to act on it, like Search Console, keyword data, the live results page, and your CMS, and letting it plan the steps. - Searches for "agentic seo" grew from about 480 a month in late 2025 to 880 a month from May to September 2026, according to Google Keyword Planner. - The agent handles the legwork. A person still picks the topics, holds the opinions, and approves anything that goes live. - Google's rules haven't moved. Pages produced at scale mainly to game rankings count as scaled content abuse, whoever or whatever wrote them. This post comes with receipts. It came out of an agentic SEO run on our own site this week, so instead of another definition we'll walk through what the agent did, where a person stepped in, and what we'd watch for if you try it. ## What is agentic SEO? Agentic SEO is SEO work done by an AI agent that has tools and a goal. It decides which tool to call next, reads what comes back, and adjusts the plan. A chat prompt gives you one answer to one question. An agent takes a goal like "find what our blog should cover next" and works through the steps a junior SEO would. A chat prompt is a map. An agentic setup is a driver who reads the map, reroutes around roadworks, and rings you before taking the toll road. The tools are what make it agentic. Search Console tells the agent what you already rank for, keyword data tells it what people search, the results page shows what Google rewards, and a CMS connection lets it act. Those tools usually plug in through [MCP](https://contengi.com/blog/what-is-mcp), the [open standard](https://modelcontextprotocol.io/docs/getting-started/intro) for connecting AI apps to external systems. ## How fast is interest in agentic SEO growing? Steadily, going by search demand. Google Keyword Planner shows "agentic seo" at 480 searches in October 2025, then 590 in March 2026, 720 in April, and 880 every month from May to September. A related term tells a different story. "claude code for seo" climbed to 1,300 searches in July 2026 and fell to 170 by September. Neither of these numbers are huge, but the pattern is useful: interest in the idea is holding, while interest in one specific tool spiked and cooled. ## What did our agentic SEO run look like, step by step? The goal was plain: find topics our blog doesn't cover and should. The agent ran it through the Contengi connector inside Claude. First it loaded our [brand knowledge base](https://contengi.com/blog/how-to-set-up-a-brand-knowledge-base-for-ai), so every later step knew our audience and rules. Then it pulled our full blog list, roughly 170 entries. Next came 90 days of Search Console queries, including the striking-distance terms sitting in positions 11 to 20. It ran candidate topics through Keyword Planner for volume, then compared all three sources to find terms with demand and no matching post. That comparison surfaced things a person scrolling a spreadsheet would likely miss. "content optimisation tools" sat at an average position of 1.9, landing on our [best content engineering tools](https://contengi.com/blog/best-content-engineering-tools) post, which was never written for that query. "what is mcp" showed around 49,500 monthly searches, and we had no post on it at all. For each post it then wrote, the agent checked the live results page and the questions people ask, fetched the primary sources before citing them, found internal links that resolved, checked the draft against our banned-phrase rules, and ran a link check before saving. Every link in our MCP post came back live. ## Where did a person step in? At the decisions. The agent proposed six gaps, and we dropped one of them outright. We set the format: around 1,500 words, a TL;DR up top, headings phrased as questions, and an FAQ at the bottom. We gave the go-ahead to publish. One moment stood out. The agent had flagged [llms.txt](https://contengi.com/blog/llms-txt) as a topic we'd never covered, with a note that it needed checking. When it checked, it found our [robots.txt guide](https://contengi.com/blog/robots-txt-for-ai) already had a section on llms.txt, and it said so before writing a word. The new post kept the same stance as the old one and linked to it. That's the behaviour worth building for. An agent that checks its own claims and tells you when it was wrong is worth more than one that sounds confident. ## What can go wrong with agentic SEO? Scale is the obvious risk. An agent can draft 50 posts by Friday. [Google's spam policies](https://developers.google.com/search/docs/essentials/spam-policies) describe scaled content abuse as generating many pages mainly to manipulate rankings instead of helping people, and its [guide to generative AI search](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) warns against creating separate pages for every variation of a query. The policy looks at why a page exists, whichever tool typed it. Made-up sources come next. Models fill gaps with plausible text, and a plausible statistic with a dead link does real damage to trust. The fix is mechanical: fetch every page before citing it, and check every link before publishing. Duplicates are sneakier. Our run showed several posts recorded twice in our library under the same URL. We haven't worked out yet whether those are harmless duplicate records or two versions competing for the same query, and an agent will happily add a third if nobody looks. Then there's voice. An agent without your brand rules writes like everyone else's agent. A brand knowledge base and a linter that blocks your banned phrases keep it [sounding like you](https://contengi.com/blog/how-to-prevent-ai-slop). ## What tools do you need for agentic SEO? Four pieces. Data access, meaning Search Console, keyword volumes, and the live results page. A model that can call tools, like Claude. Your brand rules written down as files the agent reads before it writes. And a way to save to your CMS with an approval step in front of anything that publishes. You can build that yourself in [Claude Code](https://contengi.com/blog/claude-code-for-marketers), which gives you full control and asks for comfort with a terminal. Or you can use a packaged connector. The run in this post used [Contengi](https://contengi.com/), which bundles the data, the brand rules, the linter, and publishing as one connector inside Claude. ## How do you start with agentic SEO on a small site? Start with one recurring job. The monthly striking-distance review is a good first one. Have the agent pull every query in positions 11 to 20 from Search Console, group them by page, and propose a specific refresh for each page. You approve the ones that make sense, and the agent republishes them at the same URL. Our piece on [automating refresh workflows](https://contengi.com/blog/what-s-the-role-of-a-content-engineer-in-automating-refresh-workflows) goes deeper on that loop. Run that for two months before you add gap finding or new drafts. You'll learn where the agent is sharp, where it guesses, and which approval steps you can't skip. ## Frequently asked questions ### Is agentic SEO the same as using AI SEO tools? Not quite. An AI SEO tool usually does one task when you ask. An agentic setup chains tasks together, chooses the next step itself, and works from your own data. ### Will Google penalise agentic SEO? Google judges the page itself, however it was produced. Its [guidance on AI-generated content](https://developers.google.com/search/blog/2023/02/google-search-and-ai-content) says using AI to make helpful content is fine, and using it to manipulate rankings breaks its spam policies. ### Do I need to code to do agentic SEO? Not with a connector. You add it to Claude and brief it in plain English. Building your own setup in Claude Code does need some comfort with a terminal. ### Does agentic SEO help with AI search visibility? Google says the same SEO fundamentals apply to its AI features. Track whether AI assistants mention you separately with [LLM tracking](https://contengi.com/blog/llm-tracking), because rankings and mentions don't move together. ### What should an agent never do without approval? Publish, delete, or change a live page. Keep those behind a human click.