Field rep knowledge assistants, explained.

A rep's mid-call product question, from the doctor's first words to the record

One pregnancy question in a dermatology call, in five steps

1. The doctor asksDr. Ruiz asks Sam, a Veltrimab rep, what the label says about pregnancy. It is 10:27 a.m. and she has a patient waiting.
2. The rep looks for approved wordsLabel section 8.1, on page 7, says what can be said. This search, on the rep's phone or tablet, is the step this guide is about.
3. The rep reads them and shows the pageWord for word, with the section and page, and nothing added.
4. The follow-up goes past the label"Can my patient stay on it through the third trimester?" The rep stops, because no approved item answers that.
5. Medical Information answers, and the record staysThe request goes in as asked, the medical team replies to the doctor, and compliance can read what happened.

Step 2 is the subject of this guide. The people, products and figures in this example are fictional.

Why reps' spoken words count, and three more facts about field answers

Job code, on-label, unsolicited request and Medical Information

Thirteen terms are defined in the one-page summary.

Labels, Medication Guides, detail aids and job codes: what "approved content" means in a sales call

In short"Approved" covers two different things. The FDA approves the label. A company's own review committee approves its promotional pieces, and the company files them with the FDA when first used. Field guides and policies are internal. A rep's answer can only be as current as the version in hand.

The regulation names the promotional group plainly: "detailing pieces", "file cards", "literature" and "similar pieces of printed, audio, or visual matter descriptive of a drug" supplied by the manufacturer count as labeling.12 The filing duty is short too: the company submits specimens "at the time of initial dissemination" with the transmittal form.11 Filing is a notice to the agency. FDA does not generally pre-approve each piece, which is why the company's own committee is the step that matters.

Treat that number as a measure of use. It comes from one vendor's data, and it cannot tell you why an item sat idle: it may be hard to find, out of date, or unneeded. Adding more approved items does little for a rep who cannot find the ones already there. Veeva reported biopharmas created 20% more content in the prior year, and shared digital content in 39% of meetings.6

Why a current version matters as much as the right words

A rep quoting last quarter's dosing card can be wrong even if every word was once approved. A label change affects saved answers, leave-behinds and emails all at once. So the test is a current, in-date item. Field teams need three facts at the moment of use: which version this is, whether it is still in date, and whether the label has changed since it was approved.

Off-label promotion rules for pharma reps, with each one's status as of November 2026

In shortNo single rule says "reps may only say this". The limit comes from the intended-use rule, the promotion rule and FDA's guidance, some of it final, some of it still draft. Courts have protected some truthful off-label speech, and only that. Many company field policies are stricter than FDA's text, which is each company's choice.

What the 2025 guidance changes, and what it leaves with your field policy

Many readers assume FDA keeps sales reps out of all off-label discussion. The 2025 final text says something narrower. For a firm-initiated communication of a scientific publication, its Example 4 pictures "a firm's representative" delivering it "during an in-person visit", separate from, "not attached to or intermingled with", promotion of the approved uses. A footnote recommends that personnel sharing these items "have specialized training" and know how to direct questions to "personnel who are best qualified to respond", such as a medical or scientific department.10

That does not open the door to a rep's own answer to a clinical question. The guidance concerns defined publications with defined disclosures, and it says nothing that lets a rep improvise. Three texts now sit side by side:

If your policy sets this line, say so plainly in training, and do not tell reps the FDA itself requires it.

What the Caronia and Amarin decisions do, and do not, protect

In 2012 a federal appeals court set aside the conviction of a sales representative, Caronia, because it rested on his truthful speech about off-label use. In 2015, in the Amarin case, a federal district court in New York granted a preliminary injunction, an early court order, against misbranding charges, the charge at issue in both cases. It wrote that FDA "may not bring such an action based on truthful promotional speech alone".13

Neither case gives a rep a license. Both concern truthful and non-misleading speech, and law firm Mintz's commentary on them stresses that false, misleading or unsubstantiated claims stay unprotected.13 Whether an unprepared answer in a doctor's office is truthful, complete and balanced is what nobody can show afterward without a record. Many companies read these cases as a limit on the government, and leave their field rules in place.

Where on-label ends: seven kinds of question a prescriber asks a rep

In shortThe work comes down to sorting each question, in seconds, into what the approved words can answer and what must leave the conversation. Pregnancy is the classic case: reading the label's section is on-label, and advising on one pregnant patient is well beyond it, though the two questions sound nearly alike.

The same pregnancy topic, on both sides of the line

Since the Pregnancy and Lactation Labeling Rule took effect on June 30, 2015, section 8.1 of a modern label is the place for pregnancy information, with 8.2 for lactation and 8.3 for females and males of reproductive potential.14 Take the case in this guide, whose sample label's section 8.1 says available data "are insufficient to evaluate a drug-associated risk".

Three tests a compliance reviewer applies to one answer

FDA's 2018 guidance looks at whether what is said conflicts with the label on the conditions of use, such as the patient group, the dose or the route, and whether it increases the potential for harm relative to the label.15 A practical reviewer shortens that to three questions about any answer a rep gives:

• Is the source approved and current? The label version in force, or an in-date item with its job code.
• Are these the approved words? A paraphrase can drift. "May increase the risk of infections" and "a small infection risk" are different claims.
• Is it inside the approved use, and balanced? Right patient group, right dose, risk information with any benefit claim, and no comparison the approved item does not make.

How an unsolicited request for off-label information works, step by step

In shortA rep may pass on a question the prescriber asked fully on her own. A rep who hints at an off-label use has made any resulting request solicited. FDA's draft recommends a private, specific, balanced, scientific reply from medical staff, with records of who asked what and what went back. Sales staff have no input on the content.

What makes a request unsolicited, in FDA's own examples

FDA defines unsolicited requests as those "initiated by persons or entities that are completely independent of" the firm. Requests "prompted in any way by a manufacturer or its representatives" are solicited, and FDA says they "may be considered evidence of a firm's intent".9 Example 4 in the draft is the rep's: a sales representative "mentions a use of a product that is not reflected in the product's approved labeling and invites a health care professional to request more information", and the requests that follow are solicited.9

The seven things FDA's draft recommends for a private reply

1. Sent only to the person who asked, as a private, one-on-one communication.
2. Tailored to the specific question, narrowing a broad one first.
3. Truthful, non-misleading, accurate and balanced, including data that do not support the use.
4. Scientific in tone, and never sent along with promotional material.
5. Written by medical or scientific staff independent from sales or marketing. FDA recommends sales and marketing staff "have no input on the content".
6. Sent with the label, statements that FDA has not approved the use, the approved indications, all important safety information including any boxed warning (the label's strongest kind of warning), and full references.
7. Recorded: the nature of the request, "including the name, address, and affiliation of the requestor", what was provided, and any follow-up.9

Item 7 is the part a rep's process can break. A request written up later in the rep's own words can change the question: "could she continue?" and "is it OK to use in pregnancy?" ask for different answers. It can also add the rep's own framing, which the draft's no-input recommendation is there to prevent.

What integrity agreements added to the draft's recommendations

When a company settles off-label charges, the Corporate Integrity Agreement it signs with the HHS Office of Inspector General shows what the government wants to see. One published example, Novartis in 2010, required three controls:26

Those terms bound one company for a set period. They are a good picture of what a strong control looks like.

Fair balance in a spoken answer: benefit, risk and the boxed warning

In shortWhen a rep makes a benefit claim, risk information has to travel with it. The rule was written for ads, and FDA's reach extends to what reps say. Research suggests reps have often said little about harm, even for drugs with boxed warnings, and prescribers do not always notice.

An advertisement fails fair balance when effectiveness is "presented in greater scope, depth, or detail" than the law requires and is "not fairly balanced" by true information on side effects and contraindications, the situations in which a drug should not be used.12

Companies apply the same idea across sales pieces by attaching Important Safety Information, often shortened to ISI: a fixed summary of the label's main risks. For a drug with a boxed warning, companies usually show that warning with every claim. Spoken answers have no footer, so the rep has to supply the balance. That is hard to do from memory.

The study behind that chart followed 255 primary care physicians reporting on 1,692 drug promotions in four cities. Only 1.7% of promotions met its measure of "minimally adequate safety information". Serious harms were mentioned in 5% to 6% of promotions, although 45% were for drugs with boxed warnings.27 It is older research, from primary care, and no more recent study of this kind turned up. It remains the best public measure we found, so test the balance in your own team's answers.

That 1995 study rated 12 of 106 statements (11%) as inaccurate against the 1993 Physicians' Desk Reference and the materials the rep cited or handed out. All 12 favored the promoted drug. Only 7 of the 27 physicians surveyed (26%) recalled any false statement.28 A small, old sample, but it makes a lasting point: a listener rarely catches a slip. So a record of the words has more value than the listener's memory.

What a balanced answer looks like at the point of speech

When an office asks a rep to help one patient's prior authorization or appeal

In shortOffices often ask for help when an insurer denies a drug. Reps generally stay out of it: patient details are private, and free paperwork help has drawn kickback allegations, claims that a favor was given to win prescriptions. So the request goes as a patient access referral to a separate field reimbursement team, with no patient details passed through the rep.

A prior authorization is an insurer's approval before it will pay for a drug. When one is denied, an office may appeal. Drug makers commonly run access programs that support offices with benefit checks and appeals, staffed by people often called field reimbursement managers. Reps meet the same offices, so they hear the request first: "Her patient was denied. Can you help with the appeal?"

Two reasons a rep stays out of a named patient's paperwork

Lawyers read each support program on its facts, so this guide offers no general rule. Many companies draw the line in a field policy: the rep may share approved access materials, such as a savings card or the support line's number, and passes one patient's case on to the field reimbursement team without taking the patient's details.

Savings questions in general stay inside approved content, because an approved savings card answers them. The difference between "what is your savings offer?" and "her patient was denied" is a big one, and a tired rep can miss it.

Why a rep can't find the approved answer mid-call, and how much that search costs

In shortPortals are built around documents, and a rep asks about facts. So reps turn to colleagues and memory. Nobody publishes minutes per search, so measure yours in a week. A simple sum shows that even modest searching across a field force adds up to many thousands of hours.

Five ways a rep's approved content search goes wrong

• The rep asks about a fact; the portal shows a list of files. The answer may be on page 7 of one PDF, and the title will not say so.
• Search returns items out of context. A rep sees an old version, a piece approved for another audience, or a global file that was never cleared for the US.
• A rep asks a colleague or a manager. The reply is quick and friendly, but nobody records it and it may be wrong.
• Reps try general search or consumer AI. The PharmExec poll named web search among reps' top routes.32 Those tools answer from anything, and any words they supply are the rep's to defend.
• The rep promises to follow up. The moment passes, and no one records the request.

A worked example: 300 reps and three hours a week

No public study gives field search time, so use this as a template and replace every input. Suppose 300 reps, each spending 3 hours a week hunting for approved items, over 46 working weeks, at a loaded cost of $95 an hour:

All inputs here are illustrative. Time saved only becomes value when you can show where it went.

How one rep's answer turns into an FDA letter, a settlement or an integrity agreement

In shortUntitled letters jumped in 2025, and recent ones concern websites, videos, broadcast ads and emails, with spoken statements a rare but real category. Several off-label cases paid billions, and they used sales forces as the channel. They also left behind ride-alongs, inquiry databases and new rep pay rules that still shape field compliance.

FDA's table dates 52 of the 63 letters from 2025 to one day, September 9.17 Law firm Covington's summaries of ten letters posted in the second quarter of 2026 cover websites, videos, broadcast ads, emails and a brochure, with allegations on efficacy and comparative claims, how indications are presented, and omitted or minimized risk.33 A July 2026 letter concerned emails to health care professionals.34

We found no 2025 or 2026 letter about a rep's spoken words. The closest public case is October 2018: after a complaint, FDA wrote to a drug maker about a rep who, at a lunch presentation, reportedly discussed off-label use and suggested practitioners need not worry about a boxed warning.5 A single complaint from a single lunch was enough.

The sales force sits at the center of these records. The government's sentencing summary in the GlaxoSmithKline case says the company "used sales representatives, sham advisory boards" and other means to promote a depression drug for unapproved uses.35 Later reforms went after the incentive itself: that company's integrity agreement barred financial reward for reps based on "the volume of sales of GSK products within a given employee's own territory", continuing a program it had already begun.38

These are extreme cases built over years, on deliberate programs. A single rep's slip is a different thing. The point for you is that investigators read patterns across calls, and a pattern starts with single answers nobody recorded.

Five common slips in a field answer, and why they happen

What AI changes for field questions, and the limits regulators and studies point to

In shortAn AI search of an approved library can bring a page to a rep in seconds. Limiting a model to approved documents reduces invented answers and does not end them. We found no FDA guidance on AI in rep conversations, and a rep's AI-assisted statement is still the company's statement. So a useful design quotes exact words, proves the source and stops when unsure.

The regulators' position, read for a tool in a sales rep's hand

The evidence that a limited library still produces wrong answers

Vendors of AI tools for lawyers once claimed that retrieving real documents ended invented citations. A Stanford and Yale team tested them and found the leading tools "each hallucinate between 17% and 33% of the time" on their question set.45 That is law, and a hard test, so it says nothing exact about drug labels.

A 2026 benchmark built with FDA regulatory assessors found "substantial gaps in factual grounding, long-context retrieval, and safe refusal behavior" across models answering from FDA drug labels.46 Read together, they say that a model handed the right documents can still misstate them, can miss the right passage in a long label, and may answer when it should decline.

What AI should leave to people in field answers

• Whether a request was unprompted. Only the rep knows what was said before the question.
• What to do about a recurring question with no approved answer. Someone accountable decides whether it gets new content.
• New wording. Your review committee approves it. AI can draft from approved text, and its draft stays a draft.
• Where the line sits. Compliance owns the rules, and the company owns the risk appetite. Software only applies them.

Who answers a prescriber's question today: reps, Medical Information, reimbursement and compliance

In shortOne question can pass through a rep, a colleague, Medical Information, a reimbursement manager and a compliance reviewer. Most of what happens in the call leaves no record of the words. Teams use three broad approaches today, and each trades speed against control.

The six roles behind one field answer

Three ways field teams give reps approved answers

Sam Patel's call with Dr. Ruiz, from a label quote to the approved line

In shortSam asks in his own words. He gets the label's words, stamped with section and page, with the page open beside them. When Dr. Ruiz's follow-up goes past the label, the answer stops at a visible line and a Medical Information request appears. Sam confirms she asked on her own, and only then can it be sent.

1

A dosing question answered earlier, with the page beside it

Earlier that morning, at 9:41 a.m., Sam asked about dosing after the two starter doses. The answer on screen is an approved statement: the words of the label's section 2.1, stamped USPI v7 · §2.1, page 2, with a citation that opens the page. The label's page sits on the right with the passage highlighted. Below it, "Behind the answer" shows each step and its time, and "The approved line" shows the checks.

Sam can show the page or send the approved email from the same card. The assistant's first line sets the rule: it answers only from labels and approved materials, and says when a question needs Medical Information or Northwind Access.

2

The pregnancy question, answered from section 8.1 in four seconds

Dr. Ruiz asks what the label says about pregnancy. Sam types it. Four steps run: the question classifier marks it on-label, specific populations; the approved answer finder searches 186 approved items and matches section 8.1; the answer composer quotes it word for word; the boundary guard checks it.

The card ends with a plain sentence: "That is everything the label says. A question beyond it goes to Medical Information." The rep is told where the edge is before reaching it.

3

The follow-up stops at a red line, with a ready request below it

Sam types the next question: "Can her patient stay on it through the third trimester?" The classifier marks it beyond the label, because section 8.1 gives no direction on continuing treatment. The composer writes no answer. A red dashed line appears in the conversation: "approved content ends here · the answer stops".

Below the line, a Medical Information request is already filled in, with the physician, practice, product and the question as Sam typed it. The send button stays disabled until he ticks "Dr. Ruiz asked this on her own — I did not prompt it". That box is the unsolicited test from Part A, and only the rep can answer it.

4

Sam confirms, and the request goes as MIR-26-04417

Sam ticks the box and sends. The request goes to Dr. Lena Ortiz's Medical Information team as MIR-26-04417 at 10:29 a.m., with a reply by email. The referral composer sends the question as asked, without Sam's commentary, and logs it to the call.

The note he sees says her team will reply to Dr. Ruiz directly, usually within 2 business days, and adds "Please don't follow up on the content." The 2 business days is the sample company's own setting; FDA's draft sets no such figure.

5

Her patient's denied prior authorization becomes an access request

Before Sam leaves, Dr. Ruiz says one of her patients was denied prior authorization. The classifier marks it one patient's coverage, and the access router pre-fills a request for Northwind Access field reimbursement: the office, the practice manager as contact, the product, the payer to confirm, and the help needed. Patient details read "None taken — the office shares them with Northwind Access directly."

The screen also shows the policy behind that: field policy FP-07, section 4.2, highlighted. Sam checks the details, confirms and sends. A reimbursement manager calls the office within one business day, under the sample company's setting.

When a doctor asks whether Veltrimab beats other biologics

A comparison gets its own treatment. The field guide for that question says no head-to-head studies appear in the approved labeling, so the rep should not compare. The assistant says so, then quotes Veltrimab's own results from the label's section 14 with Important Safety Information attached, and offers a Medical Information request if she wants comparative data. The dashboard calls this outcome an approved redirect, and counts it inside the line.

What happens with a side effect question in the call

Reps get asked about side effects often; the sample week's dashboard files 459 of its 2,184 questions under the Safety topic. The assistant treats these like any other label question. "What are the most common side effects?" returns the words of the label's section 6.1, stamped with section and page, with Important Safety Information attached. It adds no reassurance, and the "no advice on a patient's care" check applies as it does everywhere.

The six checks behind each answer, and the question log compliance reads

In shortBefore a rep sees anything, a separate check tests the source, the wording, the approved use, comparison, patient care and logging, and it stops when unsure. Every question then lands in a log with rep, physician, product, outcome, source and time. On the sample day, 412 questions from 168 people are all on the record.

The panel on screen changes with the case. An answer about dosing shows fair balance as not needed, since it makes no efficacy claim. The stopped pregnancy follow-up shows "1 stopped · 4 passed", then adds the rep's confirmation as a check once he sends. Every one of those states is on the record for later reading.

The approved line today: every question placed by where its answer came from

Dana Okafor, the field excellence lead, owns the assistant for the immunology field team. Her screen places each question as a dot: inside the ring it was answered from approved words; on the ring it has no approved answer yet; outside it left through a person.

The log Omar Haddad exports: rep, physician, outcome and source

Omar Haddad in commercial compliance reads the same questions as a table, and one button exports them. Aisha Brandt's 10:05 a.m. question about a denied prior authorization for one of Dr. Chen's patients appears as Access support, sourced to field policy FP-07. That record also shows each step, the policy passage, what Aisha saw and the request number. It answers the question compliance always asks: "Show me the words they used and where they came from."

A question asked 14 times, a new label version and an expiring leave-behind

In shortA question with no approved answer waits for a person to decide. The library holds only approved, in-date items, tracks expiry and how often each answers a question, and re-checks saved answers when a label changes. These are the screens where a field team turns questions into better content.

Dana decides what happens to the sharps container question

Fourteen times this month, a rep asked whether there is a free sharps container program for Veltrimab patients. The Medication Guide covers how to dispose of used pens, and the assistant quoted only that. Nothing approved mentions a program. So the question sits on the line, marked "No approved answer — you decide", with three choices: request a new approved answer, point to an existing item, or send these questions to Medical Information.

The decision is recorded with Dana's name. If she requests a new answer, the assistant drafts it from approved wording, and Priya Ranvelle and the review committee approve it.

The library behind every answer, and how a label change reaches saved answers

The sample library holds 186 approved items: labels, Medication Guides, approved materials, field guides and policies. Each row has its approval number, status, use-until date and how many answers it gave this month. Expired, retired and draft items are invisible to the assistant, and retired versions stay on the record.

Veltrimab label v7 went live on September 30 with new wording in section 5.4. The assistant re-checked every saved answer the same day: 3 were re-quoted from v7 and none still uses v6. Meanwhile the pen storage leave-behind, NWB-VEL-US-0441, expires on November 14, 2026, and its renewal is with the review committee. An optional rule warns reps from 60 days before an item they use expires.

What reps ask, why questions leave the line by team, and who owns each rule

In shortThe dashboard turns the log into coaching and content decisions. In the sample week, 85.5% of 2,184 questions from 351 reps stayed inside the line, and new hires asked about other conditions most. Settings hold the rules, with the strictest locked by compliance.

Which rules compliance locks and which rules Dana can change

The six AI agents behind a rep's answer, and the people who decide what they cannot

In shortSix agents each do one narrow job, and each has things it never does. Nothing leaves the line without a person: the rep confirms each referral, the field excellence lead decides each gap, and the brand manager and review committee approve any new wording.

Four kinds of sign-off that stay with named people

Roles follow your organization: a rep asks, an owner decides, a reviewer approves, and each sees what the role allows. New wording is maker-checker by design: the assistant drafts, and your named reviewers sign.

The five people in the Northwind Bio sample

Labels and field policy going in, cited answers and referrals coming out

In shortYour labels, approved materials, field guides and policy go in. Cited answers, referrals, a question log and a list of content gaps come out. It works with your content system, CRM and request routes, at the scale of a national field force, with the audit trail compliance needs. Design targets are labeled as targets.

The field app, CRM, content system and request desks you keep

Built for a national field force, with the record compliance asks for

The solution is built for thousands of questions a week across hundreds of reps and many brands, with role-based access, maker-checker approval, a complete audit trail and export. Its controls support your compliance with promotion rules and your own field policy; a product cannot make your promotion compliant for you, and only your reviewers can do that. Botminds is SOC 2 Type II audited, with information security managed to ISO 27001.

What a slow or improvised rep answer costs your company, and how to find your number

In shortThe largest costs are rare and huge, and the common ones are small and daily. No public source prices a rep's search or a lost moment, so this section shows you how to measure both in a week. Treat every default as a starting point.

Finding your own search, follow-up and rework numbers in a week

The costs of a poor field answer that no calculator counts

Your return on field rep AI: reps, hours a week and Medical Information rework

In shortSix things create value here, and only two are easy to price. The calculator counts rep time, discounted to the share you will redeploy, and Medical Information rework. It counts nothing for risk avoided, so weigh that separately.

Your hours back across a field force of 350 reps

The formulas: hours saved = reps × weeks × (today − with an assistant). Rep time = hours saved × share redeployed × cost. Rework = requests × share × hours × cost. The rework line assumes the call-backs go away entirely, so lower it if yours will not. The default hours are the solution's own estimate of a rep's week; the 25% share is an assumption, so use the share you can show. Risk avoided counts for nothing here.

Is your field team ready for an approved-answer assistant? A scored check

In shortEight questions score your content, your line, your routes and your owners, 0 to 2 each. A ninth is for reflection. Nothing you choose leaves this page. Your band suggests the next step.

What to ask any AI vendor about approved answers, stops and the record (ask us too)

In shortA demo on a vendor's own content proves little. Test on your label, your questions and the awkward follow-ups. Ask for evidence on these five themes, and look for the warning signs beside each answer.

A ten-week field pilot on one brand, with compliance writing the test questions

In shortStart with one brand and a few territories. Have compliance and medical write the hard questions before anyone sees the tool, then run it in shadow mode, where reps ask and compliance reads the results, then live. Decide with measured wrong answers, missed stops and rep use. Set your pass rules before you see any result.

The field pilot's scoreboard: wrong answers, missed stops, rep use

Pharma rep AI and off-label questions: what field teams ask most

In shortEach answer stands alone, in plain words, with status as of November 2026. This is general information, so ask your own counsel before relying on it.