How to track AI prompts across engines
The Doily team
Reviewed by the Doily team, Oct 2026 Updated 4 min read
Quick answer
Fix a list of 20 to 30 prompts, choose the engines and locations your customers use, run the same list on a set schedule, and record mentions, citations, and competitors for every run. Judge change over several weeks, because AI answers vary from run to run.
Why it matters
AI answers are not a fixed page you can rank on. The same question can return different brands on different runs, on different engines, and in different cities. A single manual check is a sample of 1.
Tracking turns those samples into a trend. With a stable list and a steady schedule you can tell whether the work on your pages changed what engines say, or whether you are looking at normal variation.
Key takeaways
- Use one frozen prompt list, one location setting per group, and one schedule.
- Record the same fields every run so results can be compared and charted.
- Treat a single miss as noise. Read at least 4 weeks of runs before you call a change.
- Check engines separately. A good result on 1 engine says little about another.
Step by step
-
Freeze the prompt list
Use the labeled list from your prompt research, 20 to 30 prompts with group and location tags. Do not edit wording between runs. If a prompt needs to change, add the new version and keep the old one running until the overlap shows how they compare.
-
Choose the engines that matter to your customers
Start with the engines your customers talk about or use. Google says AI Overviews and AI Mode can use different techniques and show different links, so track them as separate engines. Add more engines over time rather than all at once.
-
Set the place and language
If you serve a city or region, run local prompts as that place. Keep national prompts at country level. When you check by hand, use a clean browser session and the same location every time so your own history does not shape the answer.
-
Pick a cadence you will keep
Weekly runs are enough to see direction for most small businesses. Daily runs give faster feedback while you change pages, but they show more run-to-run variation, so read them as a trend and not as a daily score. Whatever you choose, keep it constant.
-
Record the same fields every run
For each prompt and engine, log whether you were mentioned, where you appeared if the answer lists several brands, which of your URLs were cited, which competitors were named, and how the answer described you. Save the full answer text, because wording changes are useful evidence later.
-
Read trends, not single results
Compare the mention rate over 4 or more weeks per engine and per prompt group. If a prompt flips between runs, run it 3 to 5 times in a session and count how often you appear. Write down any page change you made so you can line it up with the chart.
-
Decide what to do with the gaps
Once a month, list the prompts where competitors appear and you do not. For each, check the page that should answer it, the sources the engine cited, and whether those sources say something different about you. Assign one fix per prompt and re-check on schedule.
A manual tracking sheet
Copy these columns into a spreadsheet and add one row per prompt, per engine, per run.
| Column | What to enter |
|---|---|
| Date and time | When you ran it |
| Engine | The engine name |
| Prompt and group | The exact prompt text and its group tag |
| Location | City or country used |
| Mentioned | Yes or no |
| Position | Order in the list if brands are ranked, blank otherwise |
| Cited URL | Your page that was linked, if any |
| Competitors named | Every other brand in the answer |
| Description | A short note on how the answer described you |
Add a second sheet that lists page changes with dates. Lining that list up against the mention rate is how you tell whether a change worked.
How Doily helps
Doily runs your prompt list on a schedule across up to 9 AI engines and keeps every answer, so mentions, citations, competitors, and sentiment can be read per engine and per prompt. Paid plans check 5 engines daily and Team adds all 9 weekly, the Free plan checks weekly, and the product prints the cadence next to each engine's numbers. DeepSeek results are labeled approximate.
Questions
Can I track prompts by hand?
Yes, for a small list. A sheet with 10 prompts on 2 engines is a workable start. It gets slow once you add engines, locations, and repeat runs, which is where a tracker saves the time.
Why does my result differ from what a tracker shows?
Engines can vary by run, location, and settings. A tracker records many runs under fixed settings, so it shows how often you appear rather than a single outcome.
How long before I see movement?
It depends on how fast pages are recrawled and how often engines refresh what they use. Judge a change over several weeks of runs, and do not promise a client a date.
Sources
Primary sources, checked Oct 11, 2026.
