Key takeaways
- AI has made first drafts faster to produce. Pages that include your data, customer language, and a point of view still take more time, and that's usually where teams differentiate.
- Use more AI for introductions, derivative assets, and long-tail rewrites of work you already have. Use less AI for opinion, original research, and pages that need to sound like your brand.
- A calendar, a CMS, and a chatbot don't add up to a content system. A content system maps buyer questions to specific assets and to a funnel report leadership can use.
- Buyers already ask ChatGPT and Claude. Google still sends most of the traffic for most companies. Plan for both search-engine optimization (SEO) and generative engine optimization (GEO).
- Publish fewer generic posts, and measure whether content influenced conversions. If you publish a lot of AI pages without maintaining them, traffic often rises quickly and then falls.
Your team can produce a first draft in the time it used to take to book the subject matter expert (SME). Competing teams can do the same. Leadership often still asks for more posts, because post volume is easy to report.
The work that still helps you stand out is different now. It includes giving the model information it can't invent, running publishing as a repeatable process, appearing in the places buyers already look, and measuring whether content influenced pipeline.
Most teams already use AI to write
HubSpot's 2026 State of Marketing, from more than 1,500 marketers, reports that 86.4% of marketing teams use AI in at least a few areas, and 1.7% don't use it and don't plan to. Content creation is the top use (42.5% extensively, 38% occasionally).
Widespread adoption means generating more AI drafts is unlikely to set a team apart by itself. The same report finds 73.4% of marketers see AI working alongside people, and 62.7% say we need more unique, human-centered content to compete with the amount of AI writing in the market. If the plan is to publish more using the same kind of model, that matches what most teams are already doing.
Google Search's guidance says quality matters, not how the words were produced. Using AI to mass-produce pages mainly to manipulate rankings is spam. Using AI to help make original, helpful content is allowed, and using AI doesn't give you a ranking bonus. Thin pages still tend to perform poorly. Helpful, original pages can still perform well, whether or not AI was involved.
Decide how much AI to use for each type of page
We've written about this before in when AI content is good enough. A one-line ChatGPT prompt and a workflow that includes original research and brand guidelines produce very different results, even though both get called "AI content." The second approach is much more likely to sound like your company.
AI is often sufficient for intro-level work: listicles, "what is" pages, and first-pass outlines. You can rank with that work, show up in AI search with it, and see it in pipeline. It is weaker at interesting opinion. If you ask the model for a take, it will usually summarize what already exists. Research reports, customer proof, and pages meant to change someone's mind still usually need a human editor and a human owner of the claims.
You can vary how much AI you use. Transcribing an interview is a small amount of AI. Publishing a large set of programmatic pages with a quick review is a large amount. Many useful programs sit in the middle: the model drafts, editors fix voice and facts, and the SME still owns the claim.
Content strategy should set that mix. Use AI on work that scales a library you already trust. Keep people on pages that should be written by humans.
Give the model information your team already has
The model is not strong at originality. It remixes existing material. People still produce the ideas worth using: win/loss notes, objections from sales calls, product metrics, and stories customers actually told you.
That's a common reason AI output feels generic even when the brand is strong. We've described this as a differentiation problem: if you prompt a generic model with product facts, you often get generic copy. A longer prompt usually doesn't fix that. Documented brand context does: voice, value proposition, messaging, and proof, stored where the model can use it. The open-source content system we published stores that material in a context folder: pricing, ideal customer profiles (ICPs), founder stories, and original research. It's the same information a new hire would need.
Once you have that library, you can use the model to rewrite your existing pages for additional queries. Pull long-tail queries from Google Search Console. Ask the model to answer them from your pages, in your format, then have a writer review. In that workflow, the model is drafting from a brief rather than inventing strategy.
Useful topics still sit at the intersection of what the audience cares about, what you sell, and where you have a point of view that isn't already widely published.
Use a content system, not just a publishing calendar
AI made it easier to produce more pages. Many companies responded by publishing more. That often fails if the posts aren't part of a content system.
A content system is the people, processes, and tools from ideation through measurement. SEO and GEO are modules in it, next to analytics and creation. Most teams already have pieces of this: a calendar, some dashboards, a drafting tool. What is often missing is how those pieces connect.
A calendar tells you what ships this week. A CMS stores the file. An AI drafting tool writes a first pass. None of those, on their own, answers which buyer question, at which stage, an asset serves, or how you will see it in the funnel.
The work that fills that gap is operational. Map demand from sales, customer success, and win/loss, alongside keyword tools. Give every asset a job, an owner, and a canonical URL. Run brief, SME review, edit, ship, and update as a documented process, rather than as comments scattered across Slack. Measure whether the asset did its job, rather than counting sessions alone.
We walked through that operating layer in how to tie a content system to pipeline. Content engineering supports it: large language model (LLM)-assisted research and outlines, repurposing, and quality gates. Content goes out of date, so docs-as-code habits — versioning, review, and automated checks — help you keep published pages accurate instead of leaving them unchanged after launch.
Plan for search engines and AI answers
Search now includes more than a list of blue links. Buyers use Google, ChatGPT, Perplexity, and Gemini. UserEvidence's 2025 Evidence Gap Report found 58% of B2B buyers now start software research with AI tools. 6sense's 2025 Buyer Experience Report found 94% of buyers use LLMs somewhere in the buying process.
You still need SEO and GEO. Organic search still drives the majority of traffic for most companies. LLM visibility belongs in the same plan. The rules overlap with SEO, and they aren't the same. Answer engine optimization (AEO) is the practice of writing so a model can use a section as an answer: self-contained sections, clear definitions, and FAQs that still make sense without the rest of the page.
Clicks are also harder to earn. An Ahrefs update on 300,000 keywords found that an AI Overview in the results correlated with a 58% lower average clickthrough rate for the top-ranking page (December 2023 vs. December 2025). HubSpot's 2026 report adds that updating SEO for search changes is a top trend (40.6% of marketers) and that half of Google searches include an AI overview.
Pages that repeat widely available information often get fewer clicks and fewer citations. Pages that include specific information a competitor doesn't already publish still have a role in both search results and AI answers.
Measure whether content influenced pipeline
If you can't see content-influenced leads, demos, and signups, reporting will tend to focus on posts shipped, because that number is easy to increase.
We've described reports that isolate this without a new data warehouse: CRM engagement on the URLs, de-anonymized visitors against your ideal customer profile, a Google Analytics 4 (GA4) path of "viewed this page, then converted," and grouped views of the AI-assisted set vs. the rest of the library. In one GA4 example, 3% of visitors to an AI-content page converted over three months. That's an illustration of the question you should be able to answer, not a target.
Look at traffic after launch, not just the week a page goes live. Teams that publish a large volume of thin AI pages often see a short increase, then a drop. Semrush has described that pattern as "Mount AI" in this post. Maintaining pages is part of the program. Publishing a lot of AI content without a maintenance process often produces a short-term increase followed by a decline.
In our client work, conversion rates from LLM search have run 5–10 times higher than classic organic, once tracking exists. High-intent LLM visits won't show up in a traffic dashboard if you didn't tag the referrer.
What still helps after everyone can draft with AI
Most teams can produce a fluent first draft with AI. Teams that do better tend to put original knowledge where the model can use it, decide the mix of human and AI for each type of page, run a system that maps each asset to a buyer job, and appear in search results and in AI answers. They also report content's role in the funnel, rather than only counting pages that went live in the CMS.
That takes more time than generating another listicle. It also depends on information and process that another team can't get by buying the same AI tool.
Frequently asked questions
Is AI-generated content bad for SEO?
Google doesn't treat AI as a reason to demote a page by itself. Google's guidance judges helpful, original, people-first pages, and treats scaled unhelpful pages as spam however they were made. Thin, interchangeable AI posts tend to struggle in both classic search and LLM citations. Hybrid pages with a real point of view can participate in both.
If everyone uses the same models, how is our content different?
Most teams can use similar models. What differs is the context you give them: win/loss notes, customer language, product numbers, brand voice, and a point of view that isn't already widely published. Document that, put it where the model can use it, and keep a human on claims. We walked through one version of that setup in how to create a custom GPT that knows your brand.
Should we stop publishing "what is" and listicle pages?
No. Those are a reasonable place to use more AI, and they can still rank and show up in AI search. They shouldn't crowd out the pages that carry your opinion, research, and proof. Use the intro-level pages to cover common questions, and use human-led pages when you need to be a source worth citing.
How much of the workflow should we automate?
Automate what's repetitive and rule-based: research clustering, first drafts of derivative assets, link and consistency checks, and long-tail rewrites of a library you already trust. Keep strategy, brand voice, and the claims you would defend in a sales call in human hands. Set that mix for each type of asset, rather than applying one rule to the whole library.
How do we prove this is working to leadership?
Report more than posts shipped. Pull a GA4 path of people who viewed a URL and later converted, plus CRM engagement on those URLs. Group AI-assisted pages so you can compare them to the rest of the library. If you can't answer which buyer question a piece served, and whether anyone converted after reading it, the gap is usually in how content is planned and mapped, not only in the analytics tool.
Do we still need traditional SEO if buyers ask ChatGPT?
Yes. Organic search still drives the majority of traffic for most companies, and GEO/AEO sits on top of those fundamentals rather than replacing them. UserEvidence and 6sense both show AI tools inside the buying process. The practical move is one system that publishes for search results and for AI answers.
What should we do with AI content we already published?
Keep the URLs that show up in converting sessions or warm-lead engagement. Refresh or merge the rest. Watch for a sharp rise and fall in the traffic chart (the Mount AI pattern), and don't add volume until a maintenance process exists. A smaller library that you can keep updated is usually more useful than a large set of pages nobody maintains.
Related reading
- When is AI content good enough? — Use cases where AI holds up, and where you still want a human on the page.
- How does AI change content marketing? — Discovery, measurement, and workflow shifts once generation is the easy part.
- How do you build a content system tied to pipeline, not just the blog calendar? — The operating layer that turns publishing into funnel influence.
