How To Write Blog Posts For AI Search, Interview The Client

Here is how most agencies write client blog posts today. They point an AI tool at a topic. The tool researches it. The agency publishes what comes out. It is fast, and it feels like progress. It is also the reason those posts stop working.

If you searched for how to write blog posts for AI search, here is the honest answer. The problem is not your formatting. The problem is the workflow itself. The generated post reads fine. It is long enough. It covers the topic. But it does not say one thing that twenty other posts have not already said. So an AI answer engine has no reason to quote it. This post is not another formatting checklist. It is about the step that has to come first: getting material worth quoting.

Most advice on this topic is formatting tips

Search this exact question and page one gives you the same five tips from everyone. Medium's top post says it plainly: put a summary at the top, add clear H2 and H3 tags, bold the key phrases, and number your steps. seoClarity says add short summary blocks and structure pages so machines can lift answers out. StudioHawk adds clear titles, consistent headings, tables, and schema markup. HubSpot boils it down to five bullets: cover the topic fully, cite sources, lean on experts, keep it scannable, and write like you talk.

None of that is wrong. A well-structured page is easier for an engine to read. That was true before AI search existed. It was just called good on-page SEO. Google has also said clearly that how a page gets made was never the issue. Low-value content was always the issue. Formatting does not fix low-value content.

Here is the catch. Formatting is the easiest part to automate. A model can add summary blocks and question headings to any draft, on any topic, with zero knowledge of the business. Within a year, every agency will do all five tips. Once everyone formats for extraction, formatting stops setting you apart. It becomes the price of entry.

So this post assumes you already pay that price. Formatting is half of how to write blog posts for AI search. The other half is where the words come from. We covered the technical layer in how GEO works: schema, answer paragraphs a machine can lift out, and clean site structure. Read that one for the site side. This one is about the content.

You are ranking for intent now, not keywords

None of this is new. Google has rewarded specific, first-hand content for years. Its own helpful content guidance said so long before ChatGPT existed. The pages that won hard keywords were the ones that had something nobody else had. A number nobody else published. A process nobody else explained. An opinion nobody else would say out loud.

What changed is how the machine works, not what it rewards. Old search matched your page to a phrase. AI search reads the intent behind the question, breaks it into smaller questions, and builds an answer from the best source for each piece. Generic topic coverage adds nothing to that answer. Ten other generic posts already fed the engine the same lines. The old rule did not change. The slack just ran out.

What buyers really want to know

Picture someone hiring a pool builder. They have four quotes on the table. They are trying to figure out who to trust. Under all the polite emails, they want three things. What makes this company different from the other three? How does this company actually work, from first visit to final walkthrough? And what about my specific worry: my yard, my soil, my price ceiling?

Every one of those answers lives inside the company. Not in the category. A generic post called "how to choose a pool builder" cannot say what happens when the crew hits hard rock two feet down. Only one company's real jobs can answer that. Buyers know this. That is why they ask the same specific questions on every call, no matter how many articles they have already read.

Now the important part. The AI engine wants the same thing the buyer wants. It is not looking for the best summary of pool building. It is breaking the buyer's question into pieces and hunting for the source that answers each piece best. The buyer's questions and the engine's questions are becoming the same questions. Material that satisfies one satisfies the other.

AI search turns one question into many

Here is the mechanism. A buyer asks an engine something like "who is the best pool builder near me?" The engine does not look up that phrase. Google explains that these systems pull answers together from many sources instead of pointing at one page. In practice, the engine first builds its own list of smaller questions. Then it finds the best source for each one.

For the pool builder question, that list looks something like this:

• Who shows real case studies with real project details?
• Who has reviews that sound like real customers wrote them?
• Who shares pricing, or at least explains what moves the price up or down?
• Who explains their process from the first call to the final walkthrough?
• Who has answered the exact worry this buyer has?
• Who has done this long enough to have strong opinions about materials and methods?
• Who is willing to say what they will not do, and why?

Every question on that list can be answered. And every one is its own blog post. Notice what just happened. You are not writing to rank for "pool builder" plus a city name anymore. You are writing to answer one branch of a question tree. The engine will build that tree whether you help it or not. And every branch needs answers only one company can give. A branch about case studies cannot be filled with general talk about pools. One more thing: that list is a ready-made content calendar. Every branch is a post you could write next.

Your agency cannot answer these alone

This is the part that costs me something to admit. Anyone at an agency can write a nice-sounding post about a client's business without ever talking to the client. I have done it. It reads fine. It might even rank for a while, because ranking rewards competence, not truth.

But it will be wrong in quiet ways. The process description will be close to industry standard, so nobody flags it. The pricing logic will be a smart guess dressed up as a fact. None of it will match the client's actual last three jobs. An engine reading that post finds nothing worth quoting, because everything in it was already sitting in a hundred other posts. That is not the AI's fault. The model did its job. The failure happened earlier, when nobody talked to the person who actually knows.

So the conclusion is simple, even if it stings. If the value sits inside the client's head, the job was never writing. It is extraction. The skill you need is not better prompts. It is knowing which questions to ask, and asking the person who has the answers.

The interview method

Build the question list from market research, not from a checklist you reuse on every account. Start from the fan-out. What is the market asking about this category? Where do competitors go quiet? What does the search page leave unanswered? The question list is a real deliverable on its own. It should look different for a pool builder than for an HVAC company, because their buyers ask different things.

Get the right person on the call. That means the owner, or a senior tech who does the actual work. Not whoever answers the office phone. Book thirty to forty-five minutes. Record it. Treat it as raw material for several posts, not one. A good interview always gives you more than one article can hold.

Push for specifics, and refuse the summary answer. When the client says "we use quality materials," do not move on to the next question. Ask which ones. Ask why those. Ask what the last supplier they dropped did wrong. The gap between those answers is the whole point. "We use quality materials" could describe any company. "We switched rebar suppliers after their steel came in short of spec twice" describes exactly one.

Then publish the answers in a clean structure a machine can quote. This is where the formatting tips finally earn their keep. A summary block sitting on top of real answers does real work. The sentence it sums up did not exist anywhere on the internet an hour before you wrote it.

And be honest about what does the work in this pipeline. AI runs the market research. It drafts the question list. It cleans up the transcript and assembles the first draft. What it cannot do is sit inside your client's head and produce the rebar answer. That split, AI on the pattern work and people on the judgment calls, runs the rest of a website build too. Blog writing is no exception.

When the client does not want the interview

Some clients will not give you an hour. They are busy. They do not see the point. Or they assume you already know their business well enough. Do not quietly absorb that and write the generic version anyway. Be direct.

Tell them what a post built only from public information is really competing against: every other post built from the same public information, by every other agency in the category. An engine looking at ten interchangeable posts has no reason to quote any one of them. Neither does a buyer scrolling past the tenth identical paragraph. The client is not paying for word count. They are paying for the one thing in their category nobody can copy: what they actually know. One hour on a call is the price of getting it out of their head.

How SiteWise runs this for agency partners

We built this conclusion into a service, because agencies kept hitting the same wall. As part of our SEO and GEO service, our blog content tool builds market-informed question sets for your client's exact category. It uses the same fan-out thinking this post walked through, not a generic list reused on every account.

The split of labor stays where it belongs. You run the interview, because you own the client relationship, and your client will open up to you in a way they never will to a vendor they have not met. We take the transcript and turn it into a GEO-structured post: market research, real answers, clean formatting a machine can quote.

Why bother? Because the deliverable changes. A post built on research, a real interview, and careful assembly is a different product than a generated one. Agencies that ship it charge more for it, because the client is buying research and extraction, not word count. It is also part of what makes a white-label website partner worth keeping after the first build: an ongoing content engine your clients experience as yours.

One last honest note. The production is heavily AI-supported. The substance is entirely client-sourced. The interview is the step that keeps those two facts from blurring together.