Pre-scan every memo, demand letter, NDA, discovery summary, deposition prep doc, and citation check before it reaches the supervising attorney. Sentence-level highlights show exactly which lines read AI on recitals, fact narratives, application paragraphs, and newly drafted clauses. Calibrated for the formal register and IRAC structure of paralegal work product. Aware of unauthorized practice of law limits: TextSight is a workflow hygiene tool, never a substitute for attorney review or legal advice. No training on submitted drafts. Free to try. No card.
Litigation paralegals, transactional paralegals, in-house legal-ops staff, contract administrators, and document review attorneys all face the same question by 2026: did this draft pass through an AI tool before it reached the supervising attorney, and can the firm defend the work product if a client or a court later asks. The pre-handoff scan is the cheapest answer that fits inside the supervision chain.
Paralegal work product runs from the demand letter to the privilege log narrative to the deposition prep memo. Pre-scanning fits every drafting stage because supervising attorneys, opposing counsel, partners, and sophisticated clients now read with the AI question in mind, and the supervising attorney owns the Rule 5.3 duty to direct non-lawyer staff using AI tools on a matter.
Litigation paralegals draft the first cut of motions, briefs, demand letters, and discovery responses under associate or partner supervision. AI assistance has crept into routine drafting tasks during 2024 and 2025: case summaries, fact recitations, deposition prep memos, and citation explanations. The pre-handoff scan catches AI-shaped phrasing in narrative sections before the supervising attorney reads the draft and before the partner signs anything under their name.
Transactional paralegals prepare NDA first drafts, contract addenda, closing checklists, and corporate filings. In-house legal assistants run policy summaries, contract redlines, and internal client memos that feed the GC and the business stakeholders. Both roles routinely lean on AI for stock clauses and recital sections; both roles need the residue-check before handing the draft up the supervision chain.
Document review attorneys and discovery paralegals work on Relativity, Disco, Everlaw, and Reveal platforms. The detector is not a review-platform integration; it sits next to the work product around the review itself: the deposition prep memo, the discovery summary, the privilege log narrative, and the production cover letter. Business tier API access supports lightweight integration where firms want to push drafted memos through detection before they reach the supervising attorney.
Every US jurisdiction prohibits non-attorneys from giving legal advice, accepting cases, setting fees, or signing pleadings. Paralegals work under the supervising attorney's direction, and the attorney owns Rule 5.3 supervision duties for any non-lawyer using AI tools on a matter. TextSight is a workflow hygiene tool that sits inside that supervision chain, never above it.
Unauthorized practice of law restrictions in every state prohibit non-attorneys from giving legal advice to clients, accepting cases on behalf of a firm, setting legal fees, signing pleadings, or appearing in court except where specific authorisation exists. The line is not blurred by AI tools. A paralegal who runs a scan and rewrites a sentence is doing drafting work, not advising. A paralegal who tells a client what the contract means crosses the line whether AI was involved or not. TextSight does not change the UPL framework; it operates inside it.
ABA Model Rule 5.3 makes the supervising attorney responsible for the conduct of non-lawyer assistants, including paralegals, contract reviewers, and document subscribers using AI on a matter. The attorney's duty is to ensure the assistant's work is compatible with the lawyer's own professional obligations under Rules 1.1 competence, 1.6 confidentiality, and 3.3 candor. Pre-handoff detector scanning supports the supervision chain by surfacing AI residue before the attorney's own review, which makes the attorney's Rule 5.3 oversight efficient rather than reactive.
The NALA Code of Ethics and Professional Responsibility and the NFPA Model Code of Ethics and Professional Responsibility both reinforce the same principles: paralegals work under attorney supervision, do not give legal advice, maintain client confidentiality, and disclose their non-attorney status to clients and third parties. Both codes are silent on AI tools by name, but the supervision principle covers them by extension. Certified paralegals through NALA's CP or NFPA's RP and PACE-Registered designations operate inside the same framework, and detector scanning fits without modification.
State bar paralegal regulation varies widely. California, Florida, North Carolina, and several other states have specific paralegal definitions and supervised-status requirements. Texas, Ohio, and Illinois treat paralegals as part of the supervising attorney's professional responsibility chain without a separate registration. Some states allow limited licensed paralegals (LLLT in Washington, LLP in Utah) to perform circumscribed advisory work; even there, AI tools sit inside the limited license framework. Confirm the rules in every jurisdiction where you work, and align tool use with the firm's written policy.
Detector scanning supports UPL compliance by reinforcing the supervised-drafting model: paralegal drafts, supervising attorney reviews and signs, firm bills the work. The scan surfaces AI-shaped passages before the attorney review so the attorney sees what residue, if any, made it into the draft. The professional judgment on whether the final document is fit to send remains with the supervising attorney; the scan provides the visibility, not the decision.
Pro at $19.99 a month standard, $14.99 a month on yearly billing, fits solo paralegals and legal assistants. Business at $39.99 a month, or $29.99 a month on yearly, fits in-house paralegal pods, legal-ops groups, and document review teams with 5 seats. Full details on the pricing page.
Billed $89.88/year — Save $30
Billed $179.88/year — Save $60
Billed $359.88/year — Save $120
Most paralegals see Pro pay for itself the first time a flagged memo or demand letter gets caught and rewritten upstream rather than coming back from the supervising attorney for revision. View full pricing →
Paralegal output is support drafting, not advocacy, so the editing target differs from a lawyer's brief. The job is to hand the supervising attorney a clean first cut, not to win an argument. An NDA recital, a discovery summary, and a deposition prep memo each carry a different AI risk profile, and the sentence map tells you which lines to fix before the draft goes up the chain. For the mechanics behind the score, see how AI detectors work.
The bread-and-butter paralegal task is the summary memo: digesting case law, a deposition transcript, or a discovery production into something the associate can act on in five minutes. Because the input is large and the output format is repetitive, AI assistance creeps in fast, and the memo can quietly inherit the model's hedging cadence. Since this memo becomes the associate's input rather than a finished filing, the fix is light: clean the analysis paragraphs, leave the case citations in reporter-standard form, and hand up a summary that reads like your own work.
Routine demand letters for breach of contract, unpaid invoices, or consumer disputes are commonly AI-assisted at the first-draft stage. The recitation of facts and the description of the breach often retain LLM phrasing patterns; the remedy paragraph and deadline are usually fine because the language is fixed. The narrative paragraphs are where AI residue concentrates and where the rewrite focus belongs.
NDAs are the canonical template document. Paralegals routinely use ChatGPT or in-firm AI tools to draft first versions, then revise against the firm's preferred clauses. The risk concentrates in mutual-NDA recitals, custom confidentiality definitions, and newly drafted addenda for data processing, AI use disclosure, or ESG provisions. Settled boilerplate (severability, choice of law, notice) scores low regardless and is rarely the editing target.
Discovery summaries, document production cover letters, deposition outlines, and witness prep memos are paralegal work product that sits next to the document review platform rather than inside it. AI assistance often enters at the summarisation step where input volume is large. The pre-handoff scan catches residue in the narrative summaries before the supervising attorney runs the deposition or files the cover letter.
AI-generated case citations are a known accuracy risk; even verified citations carry AI signal in the surrounding explanatory prose. Verify every citation independently against an authoritative reporter or research platform, and rewrite the explanation in your own voice. The Mata v. Avianca sanctions made the consequence of skipping citation verification stark, and the supervising attorney owns the final accuracy duty under Rule 3.3 candor.
The whole point for a paralegal is to not get the draft kicked back. The supervising attorney often runs their own detector scan, and a high score on a memo or demand letter means a revision round trip that burns a billing cycle. Scanning before the handoff means the attorney's own scan reads clean the first time and the draft moves forward instead of bouncing.
Build the first version using your firm's approved drafting tools and templates. If AI is part of the workflow under firm policy, use it as an aide for structure and first-pass prose. Always retype and edit anything from an AI suggestion so the voice lands in your own hands rather than the model's hands. Working outside the firm's stock template on new clauses or addenda is where AI residue most often leaks in.
Cite-checking is squarely paralegal work, and it is separate from any AI scan. Pull every authority on Westlaw, Lexis, Bloomberg Law, the federal reporter, or the state equivalent, and confirm the case name, the reporter, and the pinpoint resolve to a real opinion. A detector reads cadence; it cannot tell you a citation is fake. The paralegal carries the cite-check; the supervising attorney carries the final accuracy duty under Rule 3.3 when the document is signed.
Paste each section into app.textsight.ai and scan. Free handles 5,000 characters per paste; Pro handles 10,000, which covers a full demand letter, a memo section, or a long NDA clause set. For a deposition prep packet or a multi-section discovery summary, split it the way you would route it to the attorney: facts, summary, analysis, exhibits. Each scan returns an Authenticity Score and a sentence-by-sentence colour map in well under a minute.
Rewrite the sentences the map flags, keeping defined terms, matter facts, and the firm's required phrasing intact. On a memo or demand letter the work is usually a handful of narrative lines in the fact recitation and the analysis; the boilerplate clauses that flag for genre reasons are the false-positive class and you leave them alone. Because this is support drafting, the goal is a clean clear hand-up, not a polished advocacy voice. The headline score confirms the rewrite landed.
Send the revised draft to the supervising attorney for substantive review. Pro history keeps every section scan for 90 days for the matter file. PDF export gives the firm a contemporaneous record of the pre-handoff check. A typical demand letter round-trips in about fifteen minutes; a deposition prep packet in about forty-five. The attorney's own scan now reads cleanly the first time, which is where most paralegals see the tier pay for itself.
Document review on Relativity, Disco, Everlaw, and Reveal centres on responsiveness coding and privilege review, not AI detection on the produced documents. Where TextSight fits is the prose work product around the review: deposition prep, discovery summaries, privilege log narratives, and production cover letters. Business tier API supports lightweight integration for legal-ops teams pushing drafted memos through detection before they reach the supervising attorney.
Discovery summaries and production cover letters are paralegal work product that travels to opposing counsel and the court. AI assistance enters at the summarisation step because input volume is large and the output format is consistent. Pre-handoff scanning catches AI-shaped phrasing in the narrative paragraphs before the document goes out under the firm's name. The cover letter signature line is the supervising attorney's; the responsibility for the prose under it is shared.
Privilege log entries and Bates-stamp explanations need a level of precision that AI tools handle inconsistently. The privilege description column is the highest-risk field: a hedging or hallucinated description can become a privilege waiver problem under Rule 502 in federal litigation. Pre-handoff scanning catches obvious AI residue, but the privilege judgement itself is the supervising attorney's call. Detector scanning never substitutes for the privilege review.
Deposition prep memos and witness outlines are heavy AI-assistance candidates because the input includes thousands of pages of transcripts, exhibits, and prior depositions. The risk is the memo inheriting the model's hedging cadence and parallel structure. The supervising attorney walks into the deposition with the memo in hand; the memo needs to read in the attorney's voice, not the model's. Pre-scanning the memo before the attorney reads it makes the deposition prep efficient.
Business tier with REST API access supports lightweight integration where legal-ops teams want to push drafted memos through detection before they reach the supervising attorney. The audit log captures who scanned what and when, which fits the contemporaneous-record expectation under federal preservation rules. Shared scan history on 5 seats lets the supervising attorney, the lead paralegal, and the document review team see the same Authenticity Score and flagged sentences.
Sophisticated corporate and litigation clients increasingly ask, in writing, whether AI was used to draft the documents they paid for. The answer needs to be accurate and consistent across the paralegal, the supervising attorney, and the firm's billing description. Firm AI policies are real and enforced; pre-handoff scanning fits inside both the disclosure layer and the policy layer.
What the client sees is the finished document; what they do not see is the supervised drafting chain behind it. When a demand letter or memo reads AI-shaped, it tells the client the firm phoned it in, no matter how much paralegal research and attorney review actually went into the matter. The supervising attorney signs the document, but the prose under the signature came up through the paralegal first. Catching the AI residue at the handoff stage protects the firm's standing before the attorney ever has to defend it.
Most mid-market and large firms now have written AI use policies. Common patterns include: disclosure to clients when generative AI assisted a draft, outright prohibition on specific document categories such as court filings or regulatory submissions, an extra partner review step for AI-assisted drafts, and rules against including client matter details when running drafts through external AI tools. Read your firm's policy in full before relying on either AI drafting or detector scanning, and confirm with the supervising attorney on the practice group.
When a firm's client engagement requires disclosure of AI use, the person who knows whether AI touched the first draft is usually the paralegal who built it. Getting that answer right starts upstream: flag AI assistance to the supervising attorney under Rule 5.3, keep the pre-handoff scan in the matter file, and let the attorney handle the formal disclosure to the client. The 90-day history and PDF export give the firm a contemporaneous record of what was scanned and when, which is what a disclosure request actually needs.
In a paralegal pod or legal-ops group, Business tier with 5 seats and shared history lets the lead paralegal, the document review team, and the supervising attorney all read the same scan and the same flagged lines. That turns the Authenticity Score into a desk standard the whole pod drafts to, rather than a number each paralegal interprets alone. The audit log shows who scanned which section and when, which matters when a junior paralegal's discovery summary flags higher than the rest of the packet and the lead needs to see where it came from.
A paralegal does not own the Rule 1.6 confidentiality call; the supervising attorney does. TextSight is built to slot inside whatever rule the attorney sets, which usually means redacted drafts and an approved-tools list. For privileged matters, the firm's IT policy and the bar guidance where the supervising attorney is admitted are the binding context, not this page.
Whatever a paralegal pastes in, a memo, an NDA, a discovery summary, a deposition outline, is never used to train the classifier or any other model. That is written into the contract, not buried in a settings toggle, and it holds the same way on the free tier as on Pro and Business. The scan runs against the existing model and hands back a score and sentence highlights; nothing you paste feeds a training pipeline.
A practical paralegal habit is to swap party names, matter numbers, addresses, dollar figures, and any confidential facts for placeholders before pasting. The detector scores prose cadence, not who the parties are, so a redacted draft returns the same Authenticity Score as the original. On privileged matters and anything under a protective order, the confidentiality call belongs to the supervising attorney under Rule 1.6; confirm with the attorney or firm IT before any client text leaves the firm's systems.
Free scans are not kept past the result session. Pro keeps 90 days of history, which lines up with the matter file for routine drafting, and individual scans are deletable on request. Business adds configurable retention with team controls and a standard DPA on request for in-house procurement. Match the tier to the firm's external-tool retention policy and to the bar guidance in every jurisdiction where the supervising attorney practices.
Nothing here is legal advice, and a paralegal should not treat it as a green light. Before running any client matter through a third-party detector, clear it with the supervising attorney, firm risk, and firm IT. The judgments on competence, confidentiality, candor, and supervision stay with the attorney of record. TextSight is a drafting-hygiene step that lives inside the supervised chain, beneath the attorney's review, never in place of it.
More for paralegals and legal teams.
The supervising attorney's page with ABA Model Rules, Mata v. Avianca, and federal AI disclosure standing orders.
For legal writing →How a firm picks the detector its paralegal pod and attorneys share: seats, audit log, and retention.
For law firms →The integrated AI rewriter that rewrites stubborn recital and narrative passages without losing legal precision.
Read the guide →Free, Starter, Pro, Business. Yearly billing saves 25%. Business tier with 5 seats for legal-ops teams.
See pricing →Free to try. No card. Pro at $19.99/mo, or $14.99/mo on yearly. Business with 5 seats for legal-ops teams and document review groups.
How TextSight fits other teams and workflows.