COOKED
The core job is reading inbound customer letters and emails, pulling the relevant record, and composing a reply from templates or standard language — this is precisely the task current language models perform at usable quality and near-zero marginal cost. There is no license, no physical component, and no signature requirement; the discretion involved is choosing which approved response applies. Employment is already tiny (about 4,300 nationally) after decades of decline, and the remaining roles are concentrated in escalation and exception handling.
Most of this decline happened after 2021 — it is not the pandemic dip.
Median pay $38,140 → $46,800 -1.8% in real terms
This line is counted by the Bureau of Labor Statistics — the one figure on this page that isn't a judgement of ours. Headcount moves on demand, offshoring, demographics and the business cycle, and automation is one term among several, often not the loudest.
So a falling line is not evidence that AI did it, and a rising one is not evidence that it won't. Both happen in this register: some occupations resist automation and shrink anyway, others are highly automatable and keep growing. The marked year is 2020.
BLS projection, 2024–2034
-5.6%
Percentage only. The projection counts a different population from the 4,290 above — it includes self-employed workers, which for this occupation is most of them, so the two headcounts are not comparable.
Exposed, and shrinking
Both signals point the same way: the tasks are largely automatable and the BLS projects -5.6% by 2034. This is the case where the score and the forecast agree, and it is the one worth taking seriously.
Different clocks. The score is what current AI could do to this work today. The projection is how many of these jobs will exist in 2034. Everything between the two — how fast employers actually adopt, whether demand grows in the meantime — is why they can point opposite ways without either being wrong.
~700 openings a year on average, including replacing people who leave.
CorrespondentFan Mail EditorTechnical ClerkSuggestion ClerkDispute SpecialistImage Review ClerkReview CoordinatorSales CorrespondentCorrespondence ClerkOffice CorrespondentReferral CoordinatorChargeback SpecialistCorrespondence AnalystCorrespondence DictatorDocument Review AnalystAuthorization SpecialistCommercial CorrespondentMembership CorrespondentAuthorization CoordinatorCollections CorrespondentCorrespondence SpecialistLegislative CorrespondentCorrespondence CoordinatorCorrespondence Renew Clerk
Holding it up: trust premium . Weakest point: liability shield .
Core tasks are already automatable Composing a claim-denial explanation or a billing-adjustment letter from a customer's file and a library of approved paragraphs is a text-in/text-out task with the record already in the CRM — a 3 rather than a 10 because even the escalation letters follow standing language that legal or compliance pre-cleared, so there is no step in the workflow that requires leaving the screen or generating language nobody has approved before.
Fully desk- and screen-based The only physical element left is occasionally routing something to the mailroom or stuffing an enclosure; the 2 reflects that even that is print-and-fold in a climate-controlled office, and most correspondence now goes out as email or portal message.
No licence, no signature requirement No state licenses correspondence clerks, and the letters you draft go out over a department name or a manager's block signature — you are not the named party on anything, which is why this sits at 1 rather than the 5-6 a role with a voluntary certification would get.
Executes defined procedures on defined inputs You decide which approved response fits, whether a case exceeds your adjustment authority, and when to hand it to a supervisor — real calls, but the dollar thresholds and escalation triggers are written down for you, which is what separates a 3 from the 8-10 of an adjuster who sets the settlement figure.
The verdict above describes this occupation as a whole. Almost nobody does the typical version of a job — tick what's actually in your week and see how your own mix sits.
Your task mix speaks to task resistance (3/20 here) — how much of the day's work current AI already does. That is the dimension the boxes above are about.
It cannot move the other three. Liability shield (1/20) is whether the law requires a licensed human to sign. Trust premium (4/20) is whether buyers specifically pay for a person. Judgment and accountability (3/20) is whether the role exists to own consequential calls. Those are facts about the occupation's standing, not about which tasks are in your week — a paralegal who does only trial exhibits still holds no licence. Together they are 8 of this occupation's 13 points (62%).
Embodiment (2/20) is also a property of the work rather than the worker, but we don't tag individual tasks as physical or not, so the picker can't tell you anything about it. That's a limit of this tool, not a claim.
Did we get the list right? Tell us what's missing — the tasks are written from the outside, and you're reading this from the inside.
The moves above are yours to make. This is the other half: what would have to change in the world for the occupation itself to score higher. None of it is in any one person's gift, but it is where the floor actually comes from. Scores here are not a one-way ratchet. Only two of the five dimensions — task resistance and embodiment — track what machines can do. The other three track law, what buyers will pay for, and who is answerable, and those move in both directions, often in response to the same pressure AI creates. If every lever below landed, this occupation would score around 26/100, still COOKED.
Task-mix shift already underway: as templated first-response volume moves to LLM deflection, the residual desk is regulated dispute correspondence — CFPB Reg E (12 CFR 1005.11) and Reg Z (1026.13) billing-error investigations, insurance-department complaint replies under state unfair-claims-practices acts, and FOIA/records-request responses — where the reply must be grounded in a specific investigation record and mis-wording creates a regulatory finding. That residual tier is genuinely harder than the median letter.
If the surviving roles are formally redefined as complaint-resolution or grievance officers who decide the disposition (refund, denial upheld, escalation to counsel) rather than transcribe an already-made decision — the pattern in bank complaint units and state Medicaid fair-hearing correspondence — the role owns a consequential call under ambiguity rather than selecting an approved paragraph.
Human-review mandates on automated adverse decisions attaching to the notice itself: GDPR Art. 22 and EU AI Act Art. 86 (right to explanation of high-risk automated decisions), Colorado SB 24-205's consequential-decision review duty, and NYC LL 144-style audit regimes. If a named employee must attest that an adverse-action or denial letter was reviewed by a person before dispatch, the correspondence step becomes a required human sign-off — though attestation without personal liability is a weak shield.
State insurance-department rules requiring that responses to Department of Insurance consumer complaints be signed by a licensed adjuster or compliance officer, or FINRA Rule 3110(b) supervisory review of member correspondence with the public — where an existing registered principal must approve outbound letters. Expansion of principal-review requirements to AI-drafted client correspondence would pull the function inside a licensed perimeter.
The limit. No realistic route to a meaningful trust premium: buyers of correspondence are recipients, not purchasers, and no one pays extra for a human-written denial letter. Embodiment is structurally fixed at floor. Even if every lever above lands, the occupation is 4,300 people and shrinking; the levers change the character of the remaining desk more than they change its size.
| New York-Newark-Jersey City, NY-NJ | 330 | $51,650 +10% |
| Dallas-Fort Worth-Arlington, TX | 280 | $49,650 +6% |
| San Antonio-New Braunfels, TX | 130 | $48,860 +4% |
| Los Angeles-Long Beach-Anaheim, CA | 120 | — |
| Salt Lake City-Murray, UT | 90 | $47,450 +1% |
| Atlanta-Sandy Springs-Roswell, GA | 80 | — |
| Portland-Vancouver-Hillsboro, OR-WA | 80 | $65,270 +39% |
| Buffalo-Cheektowaga, NY | 60 | $48,370 +3% |
| Portland-Vancouver-Hillsboro, OR-WA | 80 | $65,270 +39% |
| Manchester-Nashua, NH | 40 | $58,270 +25% |
| Chicago-Naperville-Elgin, IL-IN | 30 | $56,030 +20% |
We have no reported case of a named organisation automating this occupation. Not one deployment, not one announcement.
That is worth saying out loud next to a score of 13. The verdict above is about what the work exposes — what current AI could do to these tasks. It is not a claim that anyone has done it. For this occupation those two things have come apart completely: the capability argument is on this page, and the evidence column is empty.
Has AI actually changed your work? One tap, anonymous, and the running tally is public. Nothing else is asked of you.
Rather than check back: get the digest and we'll tell you what changed — or watch a single occupation from its own page.