AI Blog Writing: Make the AI Ask You Questions First

Ask ChatGPT or any other AI to write a post for your shop's blog and you will have clean paragraphs in seconds. The trouble is the part that is supposed to be yours: the experience. The mistakes your customers actually make, the best time of day to come in, the real terms of the booking. The AI cannot know any of that, so it fills those gaps with something that sounds right. Your readers have no way to tell it was made up, and when you are busy, you may not catch it either.

I run a small rental photo studio in Seoul, and for a while my blog published two posts a day, written automatically around popular search terms. The posts piled up, but not one of them held a story anyone had actually lived. Visits went up; inquiries did not. So I stopped it. One thing I changed afterward was simple: before the AI writes anything, it has to send me questions first.

This post is for solo business owners and freelancers who want AI to help run their blog. Below is how I get the questions, what makes a good question, and how much I actually answer, using real questions and answers from my own studio's blog.

An illustration of a desk by a window with a laptop and a few blank cards, made with AI

AI fills the blanks with things that sound right

In early September I was preparing a post for customers who rent the studio and shoot themselves, with no photographer. The first question the AI sent me was this:

“What mistake do self-shoot customers make most often? (For example: forgetting a flash trigger / spending all their time on setup / not budgeting time for outfit changes)”

All three examples sound reasonable. They could easily happen. Had the AI written the post on its own, those three would have become “Three Self-Shoot Mistakes to Avoid.”

My answer was something else entirely:

“People stress themselves out trying to get everything perfect. Just put the camera here and there, take lots of shots, and you'll find good ones. And bring a tripod that fits your own camera.”

The mistake I see from where I stand is not a missing piece of gear. It is the mindset people walk in with. No AI, however smooth its writing, can produce that difference. Only someone who has been in the room knows it.

How I get a question card

The method is plain. Before asking for the post, I tell the AI:

“Before you write, list three to five questions about things only I would know, and show them to me first. Until I answer, don't invent any experience, numbers or reviews. Leave those spots empty.”

I call the result a “material request card.” Each card holds four things:

Why this topic — that it doesn't repeat a post I already have, and who would search for it

Three to five questions — each one answerable in a line or two

Photos needed — split into must-have and nice-to-have

A fallback direction — what to do if I can't answer, such as postponing or narrowing the topic

For the self-shoot post, the questions were: the most common mistake; whether there is a place to change and a full-length mirror; which backdrop colors the studio has right now; the best natural-light hours this season; and how many hours I recommend for a self-shoot.

Most of my answers were one line. Natural light: “9 a.m. and 5 p.m.” Hours: “The minimum booking is two hours.” I attached a few photos, and with those answers the self-shoot post went up the same day.

A white paper backdrop rolled out at Botong Lee Studio

Three rules for good questions

If the questions are sloppy, answering them feels like a chore, and you end up telling the AI to just write whatever. Three rules make answering easy.

1. Only ask what the owner alone can answer. Anything you can look up, such as regulations or general information, the AI can research itself. What it should ask you about is experience, judgment, the real terms of your service, and photos.

2. Ask so the answer fits in a line or two. “Tell me about self-shoots” is hard to answer. “The one or two mistakes you see most” gets an answer right away. Adding example options makes it even easier. But as in the case above, when your answer has nothing to do with the options, that answer is the heart of the post.

3. Decide where numbers get checked. Prices, hours and headcounts are safer not to answer from memory. While putting together the terms for my full-day rental, I found that what I remembered did not match what was actually set on my booking page. In the end I lined up the blog and my other booking channel to the booking page. Customers show up expecting the number they read there.

Answer loosely, but check the words

To the backdrop question, I wrote back roughly this: “Primary colors like red, green and yellow, and, what's it called, the one for removing the background. Also black, and a big white backdrop cloth.”

I couldn't think of the word, so I answered loosely. The AI stitched that together in its own way and wrote that the green backdrop is used for shoots where the background is removed and composited later. In reality the studio has a separate chroma key backdrop for swapping backgrounds, and red, yellow and green are simply color backdrops. I reread the post the day it went live and had the two separated.

Two lessons came out of that. You can answer the way you talk; short and loose is fine. But in the draft, check how your answer was carried over, and whether the words are the ones your customers actually use. People who shoot search for “chroma key,” not “compositing backdrop.” One word decides whether a post gets found.

The day five answers became four posts

You might expect question-driven posts to come out thin. In practice it went the other way.

When I was preparing posts on wedding ceremony photography, the AI did not start writing. Instead it explained that generic wedding-photo checklists are everywhere online, and stitching them together would produce a post that reads like every other one, so it needed to know how I actually shoot. It sent five questions.

One of them was: “Among the conditions that differ between wedding halls, what affects the photos most?” My answer: a dark hall looks lovely in person because the spotlight gathers everything, but what matters is whether the lighting follows the couple as they enter and move. When the lighting can't keep up, the photos suffer, and that is a lighting problem, not the photographer's fault. In that respect, a bright hall is the safer choice.

That single answer became a whole post on wedding hall lighting. My answer to the question about one versus two photographers became another post, and two sets of photos from friends' weddings, shared with their consent, became two portfolio posts. Four posts went up that day.

On days with no material, it doesn't write

Along with the question cards, I set one rule. If a post can stand on facts already confirmed, it gets written. If the backbone of the post is something only I know and I haven't answered yet, nothing gets written that day.

That means some days nothing goes out. That's fine. An AI will always produce something, even when it has nothing to say. Usually it's a generic piece you've read a dozen times, one that tells customers nothing about your business. A day that leaves behind a single question card is better than a blog filling up with those.

A sequence you can try today

Pick one question customers often ask you, and make it the topic.

Tell the AI: “Before you write, give me three to five questions only I can answer. Until I answer, don't make up any experience, numbers or reviews.”

Answer each in a line or two. Attach only photos you are comfortable making public, and if people appear in them, only photos they agreed to share.

In the draft, check just three things: do the numbers match your booking page, are these the words your customers use, and is there anything in it you never said?

Keep every question and answer in one file. Next time, you won't have to answer the same thing twice.

For the step after that, turning an AI draft into something that is truly yours, see how to turn an AI draft into your own work.

In short

AI fills in experience and numbers it doesn't have. To stop that, have it ask before it writes.

Good questions are ones only the owner can answer, in a line or two. When your answer ignores the example options, that's the core of the post.

Answer numbers from your booking page and contracts, not from memory.

Answer loosely if you like, but check how your words were carried over in the draft.

On days with nothing confirmed to say, let it write nothing. That is how a blog stays about your business.