Hand AI a raw contact list and get back verified titles, current employers, clean industry tags, and a source-backed confidence score on every row.
Tools required:
- An AI chat tool with web access (Claude or ChatGPT) for the manual version.
- For the automated version: a Google Sheet plus an AI model with web search, connected by Zapier or Make (or a light script) to move rows and write results back.
Estimated time: 30 minutes for a manual batch / a few hours to set up the automated version.
Complexity: Low (manual) / Medium to high (automated, mostly because of the write-back).
What it is/Overview
Every contact list rots. People change jobs, and the titles and industry tags you saved are now out of date or missing. You know the list is stale, and checking it by hand is the job no one volunteers for.
This automation puts an AI on each contact. It confirms whether the person is still at the company, finds their current title, tags the industry from a list you control, and attaches a source link plus a confidence rating for every claim.
The manual version is a prompt you paste a batch of rows into. The automated version reads your sheet and writes the enriched fields back, row by row.
Why someone should implement it
Stale data quietly taxes everything downstream. You email people who left and mis-segment the ones who stayed, so your outreach sags for reasons that have nothing to do with your message.
Manual enrichment fixes that, in theory. In practice it's slow enough that it never happens, so the list keeps decaying. Handing the first research pass to an AI gets you a checkable answer in seconds per contact, which is the difference between "someday" and done.
Expected results/outcome(s)
- Every row tagged with a current title and an industry pulled from your own list.
- Contacts who moved on flagged before you waste a campaign on them.
- A source link and confidence score on each row, so you can trust the sheet without re-checking all of it.
Step 1: Decide your fields and your industry list.
Write down the columns you want back. A solid default is current title, still-at-company, current employer if they moved, industry, sub-vertical, confidence, source URL, and notes.
The industry field is the one to lock down. Give the AI a fixed list of industry values to choose from rather than letting it invent labels, or you'll get "SaaS," "Software," and "Tech" scattered across rows that should all match. A closed list keeps the output segmentable.
Step 2: Build the enrichment prompt with two honesty rules.
Wrap your fields in a reusable prompt. Two instructions matter more than all the others, because they're what keep the AI from confidently making things up.
"For each contact I paste below, research and return: current title, whether they are still at the listed company, current employer if they moved, industry (choose only from this list: [your list]), a confidence level, a source URL, and a short note. Rule 1: every claim must have a source URL you actually found. Rule 2: if you cannot confirm the specific person, return 'Unknown' rather than guessing. Never invent a title or a source you did not actually find."
Rule 2 is the one people forget, and it's the one that saves you. When two people share a name, an unguarded AI will happily pick the wrong one and sound certain. Forcing an "Unknown" turns a confident error into an honest gap you can follow up on.
Step 3: Run a batch by hand and spot-check.
Paste 20 or 30 rows under your prompt and read what comes back. Before you trust it on the whole list, open the source links on about 10 rows and confirm they say what the AI claims.
You're looking for one failure in particular: the right name attached to the wrong person. If the confidence scores line up with what you find in the sources, and the "Unknown" rows really are ambiguous, your prompt is working. If not, tighten the wording and run the batch again.
Step 4 (optional): Automate with write-back.
Once the manual version earns its keep, wire it to run against a sheet.
What you'll need: a Google Sheet as the source of truth and an AI model with web search for the research, wired together by a connector like Zapier or Make that reads pending rows and writes results back into the same sheet.
Basic flow: the automation works through the rows that have a company but no confidence score yet, and writes each enriched result back in place. Because the sheet itself tracks what's done, a run that stops partway just picks up where it left off next time.
One hard-won caution: dedupe the list first. Real exports are full of the same person twice, and a row-matching automation can send both copies to the first occurrence and leave the rest blank. Clean that up before you point anything automated at the sheet.
Summary
Prove it on a batch by hand before you automate the write-back, and keep the two honesty rules in the prompt the whole way. The manual version pays for itself the first time it flags a contact who left 6 months ago, and the automated version keeps a whole list current with almost no ongoing effort.
If you've got a list worth cleaning up and would rather have this built to run on its own against your CRM, that's the kind of automation we put together at Champion AI. Book a quick call to see if we're a fit.
Back