Sourced people research

Confirm the subject before you take a note

Writers, analysts, producers, and researchers share a failure mode: the name is in the brief, the deadline is real, and the first plausible page becomes “the person.” A sourced people-research workflow treats that as the moment you stop, not the moment you start typing. Confirm the subject. Then collect claims that still have URLs on them. Then decide whether you are booking, quoting, or mapping a field. Those are different jobs. None of them is a background check.

OneSearch is a private public-web brief after you pick the person. It is not a newsroom CMS, not a document store for leaked files, and not a consumer reporting agency. The method on this page is usable without the product. The product is useful when the name is common and you need the citations to stay attached while you ask follow-ups. The shared sequence is in people search, without a dossier dump and how it works.

If your assignment is “find dirt” or “decide if we can hire / house / insure them,” stop. That is not this page. Use counsel and, where the law requires it, a CRA. The fork is below.

The source stack — open pages only, in order

Volume is not a method. Order is. Work from pages that are hard to fake and easy to date, toward pages that are easy to mash. If you start in press, you will inherit someone else’s homonym.

Company site or team page first. It is dull and it usually names the seat. Talks, podcasts, and bylines second — they are things the person put in the world with their name on the marquee. Public social and public code third, and only the accounts that agree with the team page or the byline. Press last. Press is where two Maria Garcias become one paragraph.

“Open pages only” means you can hand the URL to an editor or a partner and say this is where the sentence came from. It does not mean a data-broker PDF, a leaked dump, a friend-of-a-friend forward, or a directory field with no provenance. People directories sell a different object. Do not launder that object into a reported claim.

When two layers disagree, keep both lines and both dates. A 2024 team page and a 2026 podcast bumper can disagree on title. Your job is to show the disagreement, not to pick the more interesting one. The source-verification checklist is the habit: open the original, check the date, check the name against the confirm artifact, then and only then use the line.

Citations or it did not happen

A claim without a URL is a rumor you typed. That is fine in a private hypothesis list. It is not fine in a memo, a script, a briefing note for a host, or a paragraph you are about to publish. Write the sentence, paste the URL, date the access. If you cannot paste a URL, the sentence goes in a questions list, not in the brief.

Secondary summaries are not citations. “According to a profile in…” is a citation of the profile, not of the underlying talk. If the profile got the year wrong — they do — you just shipped the error. Open the talk. Cite the talk. Mention the profile only if you are writing about the profile.

Thin beats invented here more than anywhere. A four-line brief with four URLs is a document another researcher can audit. A two-page narrative with “the public record suggests” and no links is a liability. If you are embarrassed by how short the page is, you are closer to honest than the narrative is.

Worked example: Alex Chen, guest or source

An editor says: “Get Alex Chen for the latency piece — the payments engineer who did that talk last year.” That is a booking ticket and a quoting ticket at once. Split them.

Confirm as if you will publish the name

You will. So treat confirm like a legal spelling. Search the talk first, not the name. Open the video or the deck. Capture: title, event, year, employer on the bumper, face. Then open the employer team page and match the face or the middle initial. Only then write “Alex Chen, staff engineer, [employer], speaker on [talk], [year].” The NYC quant with a faculty PDF on a different kind of latency is a different person. The Austin founder with a podcast is a different person. If you book the founder because the calendar was easier, you have booked the wrong show.

Quoting

If you only need a quote from the existing talk, you may not need a new interview. Cite the timestamp. Do not paraphrase a slide you did not open. If a blog summarized the talk and the summary is cleaner than the talk, cite the talk anyway and write the cleaner sentence as your prose, not as their words.

Booking

If you need them on the record next week, the brief needs a current seat, a current contact path they have made public (work email pattern, a form, a social account that is clearly theirs), and a one-line reason that matches the talk they already gave. Do not pull a personal mobile from a directory and call it research. Do not use a people-search product to decide whether they are “safe” to book. Safety in a talent or legal sense is a different process.

Worked example: Maria Garcia, subject of a memo

An analyst is writing a field memo on municipal data vendors. A customer mentioned Maria Garcia as the person who “runs the Denver conversation.” That phrase will not survive contact with the web. You need to know whether Maria is director of data, director of partnerships, a city employee, or a journalist who covers the beat — four different memos.

Start with the vendor the customer named, not with “Maria Garcia Denver.” Team page. Conference program. A byline. If the Tuesday municipal-data panel lists her with this vendor, you have an artifact. If a Denver newspaper has a Maria Garcia who sits on a board, that is a second card until a unique fact ties them. Do not merge “Denver” + “data” + “Garcia” into one subject. The memo should say which Maria, with URLs, or it should say the customer mention could not be confirmed.

Unconfirmed mentions belong in a questions appendix: “Customer described a Maria Garcia who runs the Denver conversation — not matched to a public page as of [date].” That sentence is more useful to the next analyst than a confident wrong title. Confidence is not a source.

If you later interview her, the memo’s public-web section stays public-web. Interview notes are labeled as interview notes, dated, with your name. Do not feed those notes back into a people-search tool and then cite the tool as if it had discovered them.

Worked example: Fatima Hassan, guest vs paper

An editor says: “Get Fatima Hassan for the public-health data piece — she published that lab paper last year.” Two Fatima Hassans will fill the first page: a journalist who covers health policy, and an epidemiologist who publishes under a lab name. Booking the journalist because her email is easier is a wrong show. Quoting the journalist’s column as if it were the lab paper is a correction waiting to happen.

Search the paper title first. Open the PDF. Capture authors, lab, year, corresponding org. Then open the lab or university staff page and match the name. The journalist card is a different person until a unique join key says otherwise — it almost never will. If the assignment was actually “the reporter who wrote about the lab,” that is a different ticket; write it that way.

If you are quoting: cite the PDF and the page or timestamp, affiliation as of the paper. If you are booking: current seat as of the staff page, a contact path they made public, a one-line reason that matches the paper. Do not pull a personal mobile. Do not merge the journalist’s clips onto the epidemiologist’s card because both strings contain “health.”

Homonyms in press are the default

Trade press, local news, and AI-written roundups collide names on purpose or by accident. “Alex Chen, who has worked in payments and spoke about systems” can describe three people. Your job is to refuse the merge.

A press hit is usable when it contains a unique join key you already confirmed: this employer, this talk title, this middle initial, this photograph that matches the team page. A press hit that only contains the name and a city is not usable. File it under “possible, not attached.”

Watch for recycled bios. A speaker bureau page copies last year’s title. A conference republishes a bio from a different Alex. A LinkedIn-public extract on a third-party site lags a year. Date every reuse. If you cannot find a primary page newer than the reuse, write the claim as stale.

Academic and government PDFs are useful and dangerous. They often carry middle initials and org names — good for confirm — and they linger after people move. A faculty PDF is a card, not a current seat. Mark it that way.

Booking vs quoting vs mapping

Booking a guest: you need a confirmed person, a current affiliation, a public reason they are the right guest, and a contact path they offered the world. You do not need a life dossier. You do not need to surprise them with a fact from a people directory.

Quoting a public talk: you need the file, the timestamp, the correct spelling, and the affiliation as of the talk (which may not be current). Say “then at [employer]” if they have moved. Do not silently update the affiliation to make the sentence tidier.

Mapping a field: you need many thin, confirmed cards, not one deep dump. A landscape memo that gets three Marias right is better than a narrative that merges them. Repeat the confirm gate for every name. Copy-paste research is how fields get fictional protagonists.

These jobs can sit in the same week. They should not sit in the same paragraph without labels. “We booked Maria Garcia (director of data, vendor, team page accessed 2026-09-20) and we quote her Tuesday panel (program PDF, 2026).” Two citations, two jobs, one person — because you confirmed once and reused the card on purpose.

The eligibility fork

Researchers get asked to “just look someone up” before a hire, a fellowship, a rental, a grant, or a board seat. Some of that is ordinary due diligence on public professional pages. Some of it is a consumer report with the serial numbers filed off. You need a fork, not a vibe.

If the output will inform hiring, tenant, credit, or insurance eligibility, stop and use a compliant process. OneSearch is public-web research, not a CRA, not FCRA-compliant. Pasting a sourced brief into a “why we declined” file does not make it less like a report. Read FCRA and legal use and people search is not a background check. The statute is on the FTC Fair Credit Reporting Act page. This is not legal advice.

If the output is conversation prep, a booking brief, or a cited memo, stay in the lane: confirm, open pages, keep URLs, refuse directory filler. Do not add criminal-record flavor from a search-result snippet. If a public court page is material to a reported story, that is journalism with editors and lawyers — not a people-search sidebar.

Quiet subjects are still subjects

Many people you will need to describe barely exist on the open web. Operators, civil servants, researchers who publish under a lab name, people who share a name with a celebrity in another field. A quiet result is a result. It is not a prompt to buy a dossier so the page looks like the loud subjects.

Add breadcrumbs: org, city, event, a paper title, a docket number if the assignment is already in public records. Search the org’s staff page directly. Search the event program. If nothing attaches, write “public web quiet as of [date], confirm artifact missing” and take it to a human — an editor, a PI, the person who assigned the name. The quiet-web guide is the stop line.

Do not fill silence with a directory age, a possible relative, or a map pin. Those fields make a quiet person look documented. They are usually about someone else or about no one in particular. Quiet plus invented is the worst brief you can file.

Leave a paper trail someone else can audit

The test of a research brief is whether a colleague can reopen it next week and see how you got there. That means: the confirm artifact at the top, the date you confirmed, the URLs under each line, the lines you refused and why (“press hit, name + city only, not attached”). A chat transcript with no links is not a trail.

OneSearch keeps citations on the page and keeps follow-ups on the same confirmed person so the trail does not fork. If you work by hand, one note per confirmed subject. Do not keep a running “Garcias” doc. That document will merge people by the end of the month.

Corrections: when a subject emails to say you have the wrong affiliation, open the new primary page, date it, and amend. Do not argue from a stale conference bio. If they want a public page about them reviewed or removed from OneSearch results, that is opt out, not an edit war in your memo.

A desk method

Run this in a plain document. Use a product when the name collides and you are out of time. The gate does not move.

1. Write the assignment type: book, quote, map, or memo. Write the sentence “this is not a screen.”
2. Write the name and every breadcrumb the assignment already gave you. Do not search the bare name first.
3. Confirm with three signals (org or seat, place or unique artifact, a page you opened). If two cards fit, stop. Ask the assigner.
4. Walk the source stack in order. Date every claim. Keep disagreements.
5. Put anything without a URL in a questions list. Do not promote it.
6. Label interview notes as interview notes. Never re-cite them as web sources.
7. If someone asks you to use the brief for eligibility, refuse and send FCRA and legal use.

Recruiters running a confirm-first pass before outreach should use the recruiting brief — same stack, different first note, same eligibility stop. Salespeople who need a call script rather than a citation trail should use the first-call brief. Founders briefing an investor or partner should use the founder brief. For directory-shaped products that look like research and are not, stay on OneSearch vs people directories.