Hire a Research Integrity Analyst Who Treats a Flag as a Lead

What Does a Research Integrity Analyst Do After the Detector Fires?

A screening tool returns eleven flags on a Tuesday batch: two duplicated western blot panels, a reference list where four DOIs resolve to nothing, and five manuscripts scored high for machine-generated text. Nobody in the queue has authority to act on any of it. That gap between the flag and the decision is the entire job, and it is why the seat exists.

The work has four repeating motions. Triage, which is deciding within minutes which flags are artifacts of the tool and which deserve an hour. Provenance, which means asking for the thing behind the claim: raw microscopy files with their metadata, the analysis script, the instrument log, the ethics approval number that can be checked with the issuing body. Correspondence, which is writing to authors and their institutions in language that asserts nothing it cannot support. And disposition, which is a recommendation with a record attached, because the recommendation will be read later by people who were not there.

What forced this into standing headcount is volume on both sides. Several major publishers now run AI-related screening on submissions as routine workflow rather than as a special investigation, with Springer Nature's Geppetto tool donated into the shared STM Integrity Hub so smaller publishers can screen too ¹. At the same time, paper mills have adopted generative tools to produce fabricated text and images that are harder to spot, at a scale peer review was never built to absorb ². The Committee on Publication Ethics describes mills as an adapting, industrialized operation, which is a polite way of saying the other side iterates against your controls ³.

One scoping decision before you write the posting. If the role also builds and tunes the screening pipeline, you are hiring closer to a research agent orchestrator and should say so, because the person who evaluates a flag and the person who ships the flagging system should not be the same person for long.

Which Tells Separate a Real Integrity Analyst From a Tool Operator?

The tell is what a candidate does with a number they cannot verify. Hand them a screening report showing a manuscript at high probability of machine-generated text and ask what happens next. A tool operator escalates. An analyst asks what the tool was trained on, whether the authors write in English as a second language, and what independent evidence exists that does not depend on the score at all.

Five things worth watching for in an hour:

• They separate the signal from the finding, unprompted. A detector output is a lead. A finding is duplicated pixels with a measured offset, a citation to a journal that stopped publishing before the cited year, or an author who cannot produce the raw file. Listen for whether the second category is where they spend their sentences.
• They have a false-positive story with a name attached to the harm. Ask about a time they were wrong, or nearly wrong, about an author. The good answers are specific and uncomfortable, and they end in a process change rather than in a shrug about tool limits.
• They ask for the raw file before they ask for an explanation. Explanations are cheap and infinitely generable. Instrument metadata, version history and analysis code are not.
• They can write the letter. Give them a scenario and twenty minutes for the first email to a corresponding author. Read it for what it asserts. Anything that states misconduct as fact rather than as a question about specific evidence is a legal problem you would be hiring.
• They know when to stop screening and start escalating. A published allegation involving funding, patients or a live grant is not a queue item. It has an institutional route, and a candidate who does not name one has never been near a real case.

The anti-tell is anyone who promises reliable detection of AI-written text. That claim does not hold, and building a screening program on it produces exactly the outcome you cannot afford: accusing a careful non-native writer while the industrialized fraud, which is edited to pass, moves through untouched.

Which Backgrounds Produce Integrity Analysts, and How Did They Get Good?

The obvious feeders are journal editorial offices, research integrity officers in university research administration, and peer review managers who already handle allegations. Those people arrive knowing the correction and retraction machinery. What they often lack is forensic technique, and that is the half you can teach fastest by hiring for it directly from somewhere else.

The less obvious backgrounds are better than their resumes suggest. Working bench scientists who have generated the exact figures under scrutiny can tell a legitimate contrast adjustment from a spliced lane, and they know which raw file should exist for each panel. Systematic reviewers and evidence synthesists spend their careers checking whether a paper says what the abstract claims, and they have already met the fabricated trial. Forensic accountants and fraud investigators bring the discipline of building a case from documents rather than from impressions. Science journalists and post-publication commenters have often done more real provenance work, publicly and under legal pressure, than most people with the title.

How the good ones got good with AI is worth asking about directly, because the honest answer is unusual in this field. The ones who are strong use the tools constantly and trust them nowhere. They will describe drafting a summary of a disputed methods section with an assistant, then finding the two claims it smoothed over. Or asking a model for the literature on an assay and checking every returned citation against a resolver, because the fabricated reference is the failure mode they now recognize on sight. That habit, generating fast and verifying line by line, is the daily motion of the job rendered in miniature, which is why the interview should ask for it rather than for opinions about AI.

Candidates who avoid the tools entirely misjudge what fraud costs to produce now. Candidates who trust the output fail expensively, because a mistaken allegation written with confident assistance is a defamation exposure with a paper trail. The person to hire uses them daily and can describe the checking without being asked. If you also need someone to design that evaluation for your whole team, that is an AI skills assessment specialist, not this seat.

Source Integrity Analysts Where Retractions Actually Get Argued

Post where the work already happens in public. The research integrity community concentrates in a small number of visible places: the Committee on Publication Ethics and its member forums, the STM association's integrity working groups, the Council of Science Editors and the European Association of Science Editors, and the post-publication comment threads on PubPeer where image and statistics sleuthing is done in the open under a real audience.

That last venue is the highest-yield and the most often skipped. A candidate with a public record of provenance work has produced, without being asked, the exact artifact you want to read before an interview: an argument built from evidence, made in front of people motivated to disagree. Read three of their threads. You will learn more than a portfolio review gives you.

Feeder employers follow the same logic. Large publisher integrity and ethics teams, university research integrity offices and offices of research compliance, contract research organizations with quality assurance functions, funder oversight bodies, and the small vendors building image-forensics and reference-checking tools, whose staff know the failure modes of the software your team will be running.

Search on the duty rather than the noun, because the title has not settled. Research integrity officer, publication ethics manager, research integrity analyst, scientific integrity specialist, image forensics analyst, editorial integrity lead and research misconduct investigator all describe overlapping seats. Screen on a written sample in every case: a redacted case memo, an allegation assessment, a published correction notice they drafted. This is a writing job that happens to involve images.

How Do You Close a Research Integrity Analyst, and Where Does the Work Sit?

Close on independence before you talk about money. The people worth hiring have all watched an integrity finding get softened because the paper came from a society's biggest contributor or the university's best-funded lab. Name the reporting line, name who can overrule the analyst, and say whether an override gets written down. A vague answer on that question loses candidates who have been burned, which is most of the experienced ones.

On compensation, be honest about what is knowable. No public wage series covers this title as of September 2026, and no reliable published salary study for it turned up while writing this, so any specific figure here would be invented rather than sourced. What you can do instead is calibrate against the bands you already run: at publishers, the seat sits near senior editorial and program management; at universities, near research compliance and sponsored programs administration; in corporate R&D, near quality assurance and regulatory affairs, which usually pay the most of the three. Ask three peer organizations directly, and expect a premium for anyone with image forensics or litigation experience, because that pool is genuinely small.

What kills the offer is predictable. A job that turns out to be clearing a queue with no authority to open an investigation. No budget for the tooling or the training that keeps the forensic skill current. A reporting line into the editorial or research function whose output is being reviewed, which every candidate recognizes instantly. And a hiring process slower than the other three offers a competent investigator is currently holding. If your governance for this sits under a broader mandate, decide early whether it reports through an AI compliance officer or stands alone.

On location, most of this work travels. Screening triage, image analysis, correspondence and case documentation are done from a laptop, and remote and hybrid postings dominate. Three parts do not travel well. Interviews with accused researchers, where an institution's procedure often specifies the format. Access to raw data held on restricted or on-premise systems, particularly clinical and patient-derived material with residency requirements. And the first months of building relationships with editors and research office staff who will need to trust this person's judgment before they will forward the difficult cases. Remote with scheduled on-site weeks, and written residency terms where a data agreement imposes them, is the shape to put in the offer.

Read the evidence >