ChatGPT prompts for a Search Console export that actually work
Seven prompts to paste with a Search Console CSV into ChatGPT or Claude, what each is good for, and the checks to run before you trust the answer.
Paste a Search Console CSV into ChatGPT or Claude, say what the columns mean, and ask a narrow question. Narrow prompts give answers you can check against the file; broad ones (“improve my SEO”) give generic advice. Below are seven prompts that stay narrow, and what to verify before you act on the result.
Which prompts are worth using?
| Prompt | Best used on | Check before acting |
|---|---|---|
| “List queries with more than 100 impressions and CTR under 1%.” | Queries export | Sort the CSV yourself and count the rows |
| “Show queries with average position between 8 and 20, sorted by impressions.” | Queries export | Matches the striking distance view |
| “Compare these two files and list the ten pages that lost the most clicks.” | Two pages exports | Both files cover the same number of days |
| “Which queries appear for more than one page?” | Query and page export | See keyword cannibalisation |
| “Group these queries by intent and name each group.” | Queries export | Groups are labels, not Google’s categories |
| “Which queries contain my brand name, and what share of clicks do they bring?” | Queries export | Give it the exact brand spelling |
| “Draft three title tags for the page with the most impressions and lowest CTR.” | Pages plus queries | Titles are suggestions; test one at a time |
How should you set the prompt up?
Start with one line of context: the site, the date range, and the meaning of each column. Then ask for output as a table with the row counts it used. If the count is lower than the file, the assistant dropped rows.
What can go wrong?
Assistants can miscount rows, misread a percentage column as a fraction, or invent a reason for a drop. Search Console exports hold at most 1,000 rows (row limits), anonymised queries are missing, and data lags by a couple of days. A prompt cannot fix any of that.
Where does a phone app fit?
Exporting is a good fit for a deliberate review. For noticing a change the same day, QueryWatch, which publishes this page, offers CSV export on the free tier and push alerts on Pro, with the pages behind a change attached. It is an independent app for iPhone and Android and does not include an AI assistant. Both store links are on the product page. For the wider question of connecting assistants to your data, read can ChatGPT or Claude read Search Console data.
Not affiliated with Google, OpenAI or Anthropic.
Questions
What is the best format to give ChatGPT Search Console data?
A CSV with one row per query or page and the columns clicks, impressions, CTR and position. Upload one report at a time, and state the date range in your prompt, because the file itself usually does not say.
Can Claude find why my Search Console clicks dropped?
It can compare two date ranges and list the queries and pages that lost the most clicks. It cannot know Google's reasons, so treat its explanation as a hypothesis to check against the data.
How many rows can I export from Search Console?
Search Console exports are limited to 1,000 rows per report, so an assistant only ever sees the top rows, not your whole long tail.
Can QueryWatch export a CSV for this?
Yes. CSV export is on the free tier, and Pro adds a comparison CSV for two periods. QueryWatch does not include an AI assistant itself.
Keep reading
Get QueryWatch
Free on iPhone and Android for every property you have, with no site limit. Open Demo Mode first if you want to look around before you connect a Google account.
iPhone needs iOS 26 or later; Android needs Android 12 or later. Free to download, with an optional Pro subscription. An independent app, not affiliated with Google.