A batch job applies one prompt to every row in your spreadsheet — hundreds, thousands, sometimes tens of thousands of times. That's the whole point: you write the instruction once and the work scales. But it cuts both ways. If your prompt is slightly off — the wrong format, a misread column, a tone that doesn't fit — that mistake doesn't happen once. It happens on every row.
The fix is simple: see the model's real output before you commit to the full run. PromptMax has this built in, and it's the single most useful habit you can develop for getting clean, import-ready output the first time.
What the preview does
In the job form there's a Test first 10 rows button. Click it and PromptMax takes the first 10 rows of your uploaded spreadsheet, sends them to the exact model you've selected with the exact prompt you've written, and shows you the model's real responses — right there in the form, before you run anything.
Two things make this more than a gimmick:
- It's the real thing. The preview uses the same model, the same prompt, and the same row formatting your full batch will use. What you see is genuinely what you'll get — not an approximation.
- It doesn't use your credits. Previewing is free, so there's no reason not to do it — and no reason not to do it again after every change.
Why this beats the alternatives
There are two other ways people try to sanity-check a prompt, and the preview is better than both:
Running a small test batch. You could run 20 rows, wait for the job to finish, and open the output file. That works, but it costs credits, it makes you wait, and you have to export and open a CSV just to read a handful of results. The preview gives you the same answer in the form, for free, with nothing to download.
Pasting a row into ChatGPT. This is the most common workaround, and it's misleading — you're testing a different model, without your system prompt applied the way the batch applies it, and without the row formatting PromptMax uses. It tells you whether the idea works, not whether your batch will work. The preview removes that gap entirely.
The key idea: the preview is a faithful dry run of your batch. If the previewed output is right, the full run will be right — because it's the same model, prompt, and formatting, just on 10 rows instead of 10,000.
How to use it
What to check in the preview
Read the 10 previewed rows against these questions — they catch the large majority of output-quality issues:
- Is the output format exactly right on every row?
- Are character or word-count limits being respected?
- Does the tone match what you asked for?
- Are the right column values being used in the right places?
- Are edge cases — blank columns, unusually short values — handled the way you intended?
Include a couple of awkward rows in the first 10 if you can — a record with missing data, an unusually long one, an edge case. The preview is most valuable when it's tested against the rows most likely to break.
The preview loop
Good batch output almost never comes from writing the perfect prompt in one go. It comes from a fast loop: preview → read → adjust → preview. Because previewing is free and immediate, you can run that loop as many times as it takes without spending anything or waiting on a full job. Most prompts are dialled in within three or four passes.
This is also the fastest way to learn what a given model responds to. Watching how the output changes when you tighten a rule or restate a format teaches you more about writing prompts than any guide can.
When to also run a small batch
For most jobs, a clean preview is enough to run with confidence. On a very large or highly varied dataset, you can add one more layer of assurance: run a small batch of 20–30 rows you've hand-picked for variety, then spot-check the output before committing to the whole file. But you'll get there faster, and cheaper, by using the preview first.