The example is made up. Grace Liu, HR manager at Fernhill Architecture in the made-up city of Lakemont, wants to post a Project Coordinator job at $50,000 to $56,000. The employers, websites and numbers are all made up, and the practice copy of Claude searches nothing.
What are the most common pay range research mistakes?
| # | Mistake | Fix |
|---|---|---|
| 1 | Not opening the sources | Click a citation and read the page, small print included |
| 2 | A vague request | Say where, what to find, how to lay it out and what to compare |
| 3 | No flags for weak numbers | Ask Claude to flag old, agency-posted and employee-reported pay |
| 4 | Treating every row as equal | Read the report's gaps before you compare |
| 5 | Comparing different jobs | Check that duties and experience level match |
| 6 | Running Research again for a quick follow-up | Turn Research off first |
| 7 | Letting Claude pick the range | Ask for talking points and say "Don't pick a new range" |
| 8 | Pasting confidential pay data | Keep your own employees' salaries out of the request |
Which mistake matters most?
1. Not opening the sources. Every range in the report has a citation, a small tag with the source. In the lesson you click the calderstudioarch citation, the preview shows the page title and site, and you click it to open the page. The range matches the report: $55,000 to $62,000. But the line under the job details says "Posted March 3, 2025. This position has been filled." The report treated it as a current competitor.
Claude read the page correctly and still missed the date. That kind of miss is easy to make and easy to catch, but only if you open the page. Do it for anything you'll bring to a pay decision. The AI citations entry explains why a citation is a place to check, not proof.
Which mistakes happen in the request?
2. A vague request. "What do project coordinators get paid?" gets a vague answer. The lesson's request names the role, the kind of employer (architecture and engineering firms), the area (within 25 miles of Lakemont), and four facts to find for each one: the posted pay range, in office, hybrid or remote, paid time off, and the benefit they lead with. It asks for a table with a source link for every fact, then gives your own draft to compare.
3. No flags for weak numbers. The request also says: tell me when something is missing, old, from a recruiting agency, or self-reported by employees instead of posted by the employer. That one sentence does most of the checking work for you. The report's gaps section came back warning about two of the four rows.
Try it free, step by step
Set a Job Post Pay Range You Can Defend: a Cited Pay Scan with Claude Research. Before you post a job, turn on Research and have Claude check the pay ranges, work setup and benefits that nearby employers publish for the same role, read the cited report, catch a closed posting… You do every step yourself in a practice copy of Claude, it checks your work as you go, and “Do it for me” finishes any step you get stuck on. Free, and nothing touches your real accounts.
Start the free lessonWhich mistakes happen while reading the report?
4. Treating every row as equal. Four employers came back, but they are not the same kind of evidence. Brandt Wells and Calder Studio post ranges on their own careers pages. Northpoint's $60,000 to $70,000 comes from a recruiting agency's post on a job board, and agency ranges can differ from the employer's. Ridgeway Partners doesn't post pay at all; the "about $52,000" comes from two employee reports on a pay-review site. Read the gaps section before you compare anything.
5. Comparing different jobs. Titles differ between firms. The report itself says to check that each job has the same duties and experience level before comparing. A "Project Coordinator" at an engineering firm may not do the same work as one at an architecture studio.
Which mistakes happen in the follow-up?
6. Running Research again for a quick question. For "give me three talking points" you don't need a whole new research run. Research usually switches itself off when the report is done. If the blue magnifier next to the + is still lit, click it to turn Research off, then click the reply box.
7. Letting Claude pick the range. The lesson's follow-up tells Claude what you found (the Calder posting is old, mark it out of date), asks for 3 short talking points for Thursday's meeting with the partners, and ends: "Don't pick a new range for us." The range depends on your budget, internal pay equity and local rules. That decision stays with you and the partners.
Claude's reply marked the old row, wrote three points (the range starts low, hybrid is common, Fernhill has an edge on leave) and added a warning: postings close and change, agency ranges can differ, and it can misread a page. Check each source again before the meeting.
8. Pasting confidential pay data. Research reads public pages. Don't paste the salaries of your own employees or anything confidential into the request. Your draft range for a new posting is fine; your payroll is not.
Key takeaways
- Open the sources. A correct range can still be from a closed posting.
- Ask for flags on old, agency-posted and employee-reported numbers in the request.
- Read the gaps and check that the jobs really match.
- Turn Research off for quick follow-ups.
- Let Claude prepare the decision, not make it. Check local pay-transparency rules before you post.
The Claude Research pay range FAQ answers the how-to questions, and Claude Research mistakes covers the same tool for other jobs. For other HR lessons, see the recruiting interview scorecard lesson.
Frequently asked questions
- Why didn't the report notice the posting was closed?
Claude read the range correctly but missed the date in the small print. Pages can be misread, which is why you open the sources yourself.
- Are employee-reported salaries useless?
No, but they are weaker than a posted range. Ridgeway's number came from just two employee reports. Keep it as a hint, not a benchmark.
- Do I need to check pay-transparency rules?
Yes. The lesson notes that several states and cities have pay-transparency rules for job posts. Check the rules where you post the job before the range goes in.
- Can I reuse the request for the next posting?
Yes. Change the role, the area and your draft. Ranges move, so the lesson suggests rerunning it before every new posting.
