Most AI ad tools pick the font for you unless you tell them, and ChatGPT works the same way: it draws the letters as part of the picture, so the font is whatever it decided looks right. That's why I describe the type in plain words, like one plain bold font, two colors at most and easy to read from across a room, and I tell it to copy every word exactly as I typed it. OpenAI's own docs say the image model can still struggle with precise text placement and clarity, so the words are the part I trust least.
If one word comes out wrong or the letters look weird, I don't start over. I open the image, use Select to highlight just that spot and type what it should say. OpenAI warns in its help article on images that highlights aren't always exact and the edit can spill past the area, so I read the whole thing again after. And if it keeps mangling the phone number, I leave the number off the picture and put it in the post text, where I type it myself. The prompts I use, with the font line already in them, are in my one photo, three ads guide.
More questions on this
Part of: Contractor Marketing, the full guide, with every related question linked.