I ran our own ad through ChatGPT. Here's the honest result.

At a Midwest Community Publications webinar, two people asked me what turned out to be the same question. A sales manager wanted to know why his proposals, sales one-sheets, and presentations come back with a different design every single time, when all he wants is the same look and feel with new content dropped in. A publisher wanted a weekly real estate ad where the footer never moves and only the listing and the photo change. One problem, one fix.

So this week I stopped theorizing and tried it, using one of our own real estate ads. ChatGPT is the tool I reach for when I want to create an image. There are others, but it is the one I know best. Here is exactly what happened, and how to get it close to exact.

What I got the first time

The first result was good, and also not right. Side by side with our real ad, three things were off: the address spacing was different, the description came out in a lighter, thinner font, and the Realtor, MLS, and Equal Housing logos were wrong, because ChatGPT tried to draw them from memory instead of using the real ones.

That last miss is the tell, and it points straight to the fix.

The fix: give ChatGPT your real ad to work from

Here is the single most important thing on this page. When you ask ChatGPT to "make an ad," it invents a brand-new one from a blank page, which is why the fonts, spacing, and logos come out almost-but-not-quite. Instead, upload your actual ad and tell it to change only the parts that change: the photo, the address, the price, and the description, and keep everything else the same.

It still rebuilds the image, so it will not be pixel-perfect, but now it is working from your real ad instead of from a blank imagination, and that alone fixes most of what I saw. The words that worked best for me were blunt: "Treat my original ad as the exact source of truth. Reproduce it and change only the listing details. Do not redesign or approximate it." The logos are the one stubborn exception, and we will handle those below.

The setup, step by step

1 Let ChatGPT write your template

You don't have to describe your ad from scratch. Open a chat, upload one finished ad you like, and have ChatGPT write the recipe for you. Paste this:

Look at the attached ad and write me a reusable template for editing it each week, in plain words. List three things:
1. What must NEVER change: the logos, the layout, the footer text word-for-word, the fonts, and the colors.
2. What DOES change each week: the property photo, the address, the price, and the description.
3. The style rules for the description: its tone, its length, and how it should read.
Write it clearly enough that a brand-new employee could follow it.

Read it over and fix anything it missed. That paragraph is your template.

2 Build a Project so you never re-explain it

A Project in ChatGPT is a folder that remembers your rules. Set it up once.

1. In the left menu, click Projects, create one, and name it (like "Weekly Real Estate Ad").
2. Open the project's Instructions and paste the template from Step 1. These rules now apply to every chat you start inside this project.
3. Click Sources, then Add Sources, and upload your real ad (the source of truth) and your logo files. Sources is just ChatGPT's word for the project's files, and a PDF is fine.

Then each week, open the Project, upload the new photo, and paste this:

Here is this week's listing. Edit our ad from the Sources and change ONLY these four things:
- The property photo (use the one I just uploaded).
- The address: [NEW ADDRESS]
- The price: [NEW PRICE]
- The description: [PASTE THE NEW LISTING DETAILS]
Keep the logos, the layout, the fonts, the colors, and the footer exactly as they are in the original.

3 Fix the drift, then teach your template

Your first result will not be perfect, and that is normal. This is the step most people skip, and it is the one that matters. Put your original ad and ChatGPT's version side by side and look for the usual drift. In my own test, these were the culprits, with the plain correction for each:

• It changed the shape. ChatGPT widened the ad toward a 4:3 rectangle, which threw off everything else. Tell it: "Keep the exact size and proportions of my original. Do not change the shape."
• The text got heavier. Our description is a light, narrow font, and ChatGPT made it bold and wide. Tell it: "The description is too bold and too wide. Match the original's lighter, narrower font."
• The address got split. Tell it: "Keep the address and town as one centered line, exactly like the original. Do not split them or add spacing."
• The photo grew. Tell it: "Keep the photo the same size as the original. Do not enlarge it."
• The copy did not fit. The right fix is to shorten the words, never to shrink the type. Tell it: "If the description does not fit, shorten the wording. Do not shrink the font or resize anything else."

Then do the part that pays off forever: take the corrections that worked and add them into your template in the project Instructions. Your template is a living document, and version one is never the final one. After two or three weeks of small fixes it settles down, and new ads start coming out right the first time.

When your rules outgrow the box

Here is a wall you will hit as your template gets good: ChatGPT limits the Project Instructions to about 8,000 characters, and a thorough set of rules runs longer than that. Do not try to cram it all in (have ChatGPT help you with it if needed). Keep a short version of the rules in the Instructions box, and put the full, detailed version in its own file in Sources, right next to your original ad. This can be saved as a .txt file.. Then add one line to the Instructions: "Treat the uploaded original ad as the exact source of truth, and follow the uploaded spec file for the details." Now the Instructions stay short, and ChatGPT still has the whole rulebook to consult every time. The one rule to keep at the very top: "Reproduce my original exactly and change only the listing details. Do not redesign or approximate it."

The one part AI still fights you on

Be honest about the logos. The Realtor, MLS, and Equal Housing marks are official trademarks with exact shapes, and no AI image tool draws them reliably. Do not let it try. Keep them from your original ad, or drop in the official files your association provides. The same goes for razor-perfect type on a big client's ad: that last five percent of exactness is still faster to finish in your layout program. For everything short of that, editing in ChatGPT gets you there.

Your one-sheets and proposals are the same problem

The sales manager at that webinar was not asking about ads at all. He wanted his proposals, one-sheets, and presentations to come back looking like his company every time, instead of a brand-new design on every request. It is the same fix, because it is the same mistake underneath. Asking AI to "make me a one-sheet" is asking it to invent one, and invention is the exact thing you do not want from a document that is supposed to look identical every week.

So run the same three steps. Take the one-sheet you are proudest of and upload it. Have ChatGPT write the template from it: what never changes (your header, your colors, your footer, your fonts, your boilerplate) and what does (the client name, the numbers, the offer). Save that in a Project called something like "Client One-Sheets," with the original file sitting in Sources. Every new one now starts from your real document instead of from a blank page.

One difference is worth knowing before you start. If your one-sheet or proposal lives in Word, PowerPoint, or Google Slides, do not ask ChatGPT for a picture of a document. Have it write the words to your template's rules and paste them into your real file, where your fonts and logos are already correct. Ask for text when the deliverable is text. Ask for an image only when the deliverable is genuinely an image, like the ad.

What this costs, and what the free plan really gets you

Everything above works on a free ChatGPT account. Projects opened up to free users in September 2025, and a free project holds five files, which covers your original ad, your logo files, and a spec file with room left over. Plus holds twenty-five, Pro and Business hold forty.

The limit that bites is not text, it is images. Everyday chatting is effectively unlimited on the free plan. Image creation is not, and OpenAI no longer publishes a firm daily number. Free users generally report about two to three images in a rolling twenty-four hour window, and you can check what you have left under Settings, then Usage.

Here is what that means for real work. Every correction round in Step 3 spends one of those images. Building your template the first time takes several rounds, so on the free plan your first ad realistically stretches across two or three days of waiting for the meter to reset. Once the template is dialed in and an ad comes back right in a try or two, one ad a week fits inside the free plan fine. What does not fit is deadline day with three ads to build.

So start free. It costs nothing to prove this works on your own ad, and proving it is the whole point. If it turns into a weekly habit, ChatGPT Go runs eight dollars a month in the US and advertises ten times the free plan's messages, file uploads, and image creation, which is the cheapest way past the only wall you are likely to hit. Plus is twenty. One thing to know before you pick: OpenAI has said ads are coming to both the free and Go plans.

Why it is still worth it

Even landing at ninety percent, you have turned an hour of rebuilding an ad into a ten-minute edit, and it improves every week as your template sharpens. That is the real promise here. Not a magic button, but a repeatable one you own and make better over time.

One last honest step: on a big client's ad, finish the type in your layout program. Two minutes there at the end of a ten-minute job still beats the hour it used to take.