The Agentic Content Operations Field Guide

You heard the talk. This is the part you run on Monday. Everything here works for a two person comms team or a twenty person marketing department, and none of it requires buying anything from us.

The whole talk in ten lines

1. Your content now serves three audiences: humans, SEO crawlers, and AEO answer engines. SEO's big brother came to town.
2. Discoverability happens twice: before anyone visits (answer engines) and after they arrive (your site's own search, nav, chat, voice).
3. Getting cited is about 70% engineering and 30% luck, and everyone is at stage one of a hundred. Distrust anyone who claims otherwise.
4. Content gets found when it is answered, fresh, and connected. Miss any one and it does not.
5. Build a Question and Answer Mesh: fifty real questions from your real data, answered in three forms, threaded through your site.
6. A piece of content is not words. It is up to twenty attributes, and the content model is the contract between your humans and your agents.
7. Governance is a production step, not a review step. Fewer humans, not no humans, and the humans do judgment.
8. Never cut and paste again. Meet your authors, editors, and publishers at their interface of choice, and automate the movement behind them.
9. Variants are an exponent (languages times segments times A/B). Exponents don't scare workflows. They crush people.
10. It takes roughly 10x the effort of one great post to build the system that produces great posts. Invest once, released for the future.

Factor 1: Discoverability

How do you find out what the robots think of you?

You ask them. Open ChatGPT, Claude, and Perplexity and ask the question you most want to be the answer to. Note who gets cited and why. Then ask the engine directly: "why did you recommend them?" The answers are revealing and sometimes humbling. Do this monthly, same questions, every engine, and screenshot the results. Visibility is per engine, not one number.

The discoverable content checklist

Under the hood

• Metadata complete on every page that matters (title, description, canonical)
• Schema applied at the template level, never typed by hand per page
• Honest expectation: schema is hygiene and entity disambiguation, not a citation lever. Controlled tests show citations barely move on schema alone. Do it anyway, cheaply.

On the page

• The answer appears in the first screen, before the story
• Question-shaped headings that match how people actually ask
• Named authors, real people, consistent across the site
• Internal links from every answer to related answers
• Clean, boring URLs

Around the web (entity presence)

• A Wikipedia presence if you plausibly merit one
• Review platforms relevant to your sector, actively maintained
• Partner and member organization links pointing at you
• Newswire or sector press for the things you do that are new
• Your organization described the same way everywhere it appears

The two-year lesson from our own experiment: the entity presence turned out to be foundational, disciplined metadata and refreshed old content moved the needle, and the engines will happily tell you they ranked you because "you say you're the best." Take the luck. Engineer the rest.

Factor 2: Content models

What makes content findable?

Answered. It responds to a real question, directly, in the first paragraph. Not eventually. First.

Fresh. Citation rates drop sharply on content older than about three months in fast-moving topics. Some of our biggest gains came from refreshing posts written three and four years ago. Old keyword-stuffed posts are space garbage: fix them, don't delete them.

Connected. Linked in, linked out, linked across. Wherever a machine lands, it should find an answer connected to more answers.

The Question and Answer Mesh, step by step

1. Gather your data. Google Search Console (what people search to find you), GA4 (what they do when they arrive), your competitors' sites, and your sector's common questions.
2. Generate fifty questions. Work with an AI assistant on this. The prompt below is the one to start from. The output is fifty questions your desired visitor (donor, member, service seeker) is highly likely to ask, ranked by how much each matters to your mission and how winnable it is.
3. Answer them in three forms. The Q&A type: ten questions answered concisely in one place. The deep dive: one question, one page, fully answered, the hub. The mesh: those answers threaded as passages and links through related content across the site.
4. Spoke it out. Your site is the hub. LinkedIn, Facebook, email, wherever you publish, those are spokes, and every spoke points home.
5. Refresh on a cadence. Every answer gets a review date. Old answers get updated, not abandoned.

The fifty questions prompt

Paste this into your AI assistant, attach or paste your Search Console export, and edit the bracketed parts:

You are helping [ORGANIZATION], a [SECTOR] organization whose desired visitors are [DONORS / MEMBERS / SERVICE SEEKERS]. Using the attached Search Console data, this list of competitor sites [LIST], and your knowledge of the sector, produce the 50 questions our desired visitors are most likely to ask an AI assistant or search engine this year. For each question: the question in natural language, who is asking it, how often it is likely asked (high/medium/low), whether our site currently answers it (yes/partially/no), and how winnable it is for us. Rank by mission impact times winnability. Do not invent search volumes. Flag any question where you are guessing.

Factor 3: Production

What is a piece of content, actually?

Not words. Attributes. A typical post carries up to twenty of them:

Title · slug · meta title · meta description · summary and answer-first opening · body · author (a real, named person) · publish date · review date · category · tags · target questions · schema type · citations and sources · hero image and alt text · internal links · outbound links · read time · channel variants (LinkedIn, Facebook, email cuts) · language variants

The shift with agents in the loop: the content model becomes the contract between your humans and your agents. It is how you tell an agent what quality means. If it is not in the model, the agent will not do it, and no one will notice until it is public.

The governance rules that saved us

• Fact gate before publish. A draft once cited a plausible statistic for a client that the AI had generated from an industry benchmark. A human caught it at the gate. Nothing with a number, a name, or a claim ships without a human pass.
• The tone filter. Hand-write exactly two pages: your homepage and one representative page. That pair sets the voice for everything the AI writes after. This is the honest answer to "how much human writing do we need?"
• The brand prompts library. Your tone prompt, governance prompt, and terminology rules become an asset. Version them, protect them, and treat them as secure as the data they understand.
• The 10x rule. Building the workflow that produces great posts costs about ten times the effort of writing one great post. Most people quit at three. Invest once and you never do that work again.

No content? Walk the building.

The lowest-tech production system we ever shipped: one marketing manager with "nothing to talk about" walks her property for one hour a week and talks into her phone. What's new, what's coming, what she noticed. Dictation goes into the pipeline, drafts come out. There are five sparks of content in every walk. Your version is the program director walking the shelter, the conference floor, the food bank.

Factor 4: Movement

Never cut and paste again. It's 2026.

The old chain: a Google Doc, fourteen editors, legal, translation, legal again, assets from a contractor, a project board holding it together, and one person at the end pasting twenty fields into a CMS with two windows open. Somebody built that system carefully. Respect it, then replace it.

The interface of choice

The future of content supply chain management is shaped by one thing people forget: the author's, editor's, and publisher's interface of choice. People have a tool they love. The pipeline's job is to meet them there.

They love	The pipeline
Word or Google Docs	A workflow tool (n8n or similar) picks content up and moves it: translation, legal, CMS
An AI assistant (Claude, etc.)	The assistant is the interface; an MCP server connects it to the CMS
The CMS itself	Authoring in place. Still fine. A choice now, not a sentence.
Their phone	Dictation in, governed pipeline out (see the walk, above)
A project board (Trello, etc.)	Card moved to "ready" triggers the pipeline
None of the above	A custom interface: analytics in, content out, one governed view

Magic versus conveyor: the honest trade-off

The MCP route is the magic version. Content in, voila, published. Wonderful for small teams. One challenge: there is no full log. It is just magically there.

The workflow route is the conveyor belt. Every node logs. Legal signs off as a node. Each language can route to the model that handles it best. Self-hostable inside your own fence. Choose this the moment compliance matters or scale arrives.

Why scale arrives sooner than you think: variants are an exponent. Nine languages makes one page nine pages. Add three audience segments and a couple of A/B cells and one page is fifty. That is why organizations buy personalization and never turn it on. The license was affordable; the variants weren't. The exponent does not scare a workflow.

Migration is movement too

Agents have changed the most feared project in the portfolio. Methodology: audit, map, model, migrate, verify, every time. Compliance: every decision logged, every page traceable. Infrastructure: the agents build the content models, components, redirects, and DevOps, not just the copy. Speed: months of work compressed into days. The strategic point: migration used to preserve your bad structure because reshaping content was the expensive part. That constraint is gone. A migration or platform upgrade is now the cheapest moment you will ever have to fix your structure for machines.

Factor 5: The future

The pieces converge: your analytics (Search Console, GA4, competitive data, citation sampling) decide what gets made; agents produce it against your content model; variants fan out automatically; a human reviews, approves, and publishes; everything is logged; and the results feed the next round. Measurement feeds production. One governed view, origin to publish.

We call our version Content Operator, and it is in beta with a co-op telco whose team's design target is about three hours a month of human time: review, approve, publish. But the pattern matters more than any product, which is why we also train teams to build this infrastructure themselves. The bottleneck this dissolves is not headcount. It is approvals per piece, and a twenty person nonprofit marketing team suffers it worse than a two person one.

And the honest question at the end of it, from a real client: if the pipeline can publish straight to components on an edge network, what do we even need a CMS for? We don't have a tidy answer. Nobody does yet.

The 90-day starter plan

Weeks 1 and 2: measure. Ask the robots your ten most important questions across three engines. Screenshot everything. Run the discoverability checklist against your top twenty pages. You now have a baseline nobody in your organization has ever seen.

Weeks 3 to 6: the mesh. Run the fifty questions prompt against your real data. Pick the ten most winnable. Write or refresh answer-first pages for them, named authors, question-shaped headings, linked to each other. Refresh beats new: fix the space garbage first.

Weeks 7 to 10: one workflow. Pick the single most repetitive movement in your content life (usually translation, channel variants, or the final paste into the CMS) and automate that one thing, using whichever route matches your team's interface of choice. Expect it to take ten times the effort of doing it once. That is normal. That is the investment.

Weeks 11 to 13: close the loop. Re-ask the robots the same ten questions. Compare screenshots. Feed what you learn into the next ten questions. You are now running agentic content operations, at whatever scale you can sustain, with a human hand on every publish button.

Budget honesty: everything above can be done with free tiers and one AI subscription. The paid version buys you speed and logging, not access to the method.

Keep in touch

• The Content Operator, our newsletter on agentic content operations. Edition one launched with this talk.
• Beyond the CMS, the podcast. If this guide sparked an idea, or you think part of it is wrong, come on the show and say so.
• cbryce@dotfusion.com · dotfusion.com

This guide was conceived and whiteboarded by Chris Bryce, with a stack of AI agents doing the heavy lifting on research and logistics. Dotfusion is a Certified B Corporation. Our first website, in 2000, was for a nonprofit. This is what we can hand one now.