A wrong number gets caught
The NBFC's interest-rate claim is held for review rather than published on a marketing team's say-so alone, giving someone with the current rate sheet a chance to catch a mismatch before it reaches a customer.
Coming soonCopywriter
Route risky claims to a human
The business avoids publishing a health, financial, or legal claim that could create regulatory or reputational trouble, because a person reviews it first.
Works with
What it does
How Lekha handles this
When a draft contains language that reads as a health outcome, financial return, guarantee, or legal assurance, Lekha automatically flags it and blocks one-click publishing until a designated reviewer approves it. The flagged sentence and the reason it was flagged are shown side by side with the draft.
A regional NBFC's marketing team drafts a loan-product social post mentioning an interest rate range, and nobody outside marketing checks that the number matches the current approved rate sheet — a single wrong digit here is not a typo, it's a claim about money that a regulator or a customer can hold the business to. This is the one capability built specifically because some mistakes are too costly to leave to a one-click publish.
When a draft contains language reading as a health outcome, a financial return, a guarantee, or a legal assurance, this automatically flags it and blocks one-click publishing until a designated reviewer approves it, showing the flagged sentence and the reason side by side with the rest of the draft, so nothing in a regulated category goes live without a person looking at it first.
Where route risky claims to a human happens
Lekha runs this directly on the platforms your customers already use — no separate app for them to install.
- WordPress
- Google Docs
- Notion
- Canva
How it works
Step by step
- 01
Every draft is scanned
A classification pass checks each sentence for language that reads as a health outcome, a financial return or guarantee, or a legal assurance, distinguishing this from safe, purely descriptive copy that carries no regulated meaning of that kind at all.
- 02
A match blocks one-click publishing
The moment a sentence is classified as a regulated-claim category, the draft is held from immediate publishing, regardless of how the rest of the piece is otherwise ready to go, until the designated reviewer has actually looked at it.
- 03
Sentence and reason, side by side
A designated reviewer sees exactly which sentence triggered the hold and why, side by side with the full draft, rather than a vague notice that something in the piece needs a second look without saying where to actually find it.
- 04
A reviewer's decision unlocks it
The assigned reviewer approves the flagged language as-is, requests a specific change, or rejects it outright, and the draft only proceeds to publishing after that explicit human decision is actually recorded against the specific flagged sentence in question.
Why it matters
What you get out of it
The specific risk, named
A reviewer sees exactly which sentence triggered the hold and the precise reason, rather than a general warning that leaves them re-reading the whole draft to find the actual problem.
Nothing regulated ships on autopilot
A health, financial, or legal claim structurally cannot reach one-click publish without an explicit human decision recorded first, closing the gap a busy day or a rushed deadline would otherwise open.
The detail
What to know before you rely on it
This exists because India's advertising and consumer-protection rules are not a footnote to marketing copy, they are binding constraints on it. The Consumer Protection Act's provisions on misleading advertisements, and the ASCI code on substantiation of claims — health outcomes, financial returns, language implying a guarantee — mean publishing an unsubstantiated claim carries real regulatory exposure, not a hypothetical one. This exists because an AI-generated sentence that reads confidently can still be exactly this kind of claim, and the model producing it has no way to verify whether the business can back it up with evidence a regulator would accept.
The detection here is a classification model, not a legal-compliance guarantee, and that matters for how this should be used. Under-flagging — missing a genuine regulated claim so it slips through — is the single highest-severity failure mode in this product, because a missed health or financial claim reaching customers is categorically worse than an unnecessary review. But over-flagging carries its own cost: if safe language gets flagged constantly alongside genuine risks, reviewers start rubber-stamping without reading carefully, quietly defeating the checkpoint's purpose. Both directions need active tuning over time.
What this explicitly does not do is decide, on its own, whether a claim is actually true or compliant — that judgment belongs to the designated human reviewer, ideally someone who can check the claim against current data, like an approved rate sheet or a licensed professional's sign-off. This is a routing and holding mechanism, ensuring that judgment call happens before publishing, not a system that replaces the judgment itself.
Industry use cases
Where this earns its keep
3 industries where Lekha applies this directly.
Lekha drafts marketing copy for financial products while automatically routing any statement about returns, rates, or guarantees to a compliance reviewer before it can be published.
A regional NBFC drafts a loan-product social post, and Lekha flags the line mentioning an interest rate range as a regulated claim, holding the post in a pending-review queue until the compliance officer confirms the number matches the current approved rate sheet.
See the banking and finance playbookLekha drafts patient-facing communication and clinic marketing copy while routing any health-outcome claim to a licensed staff member for sign-off before it goes live.
A physiotherapy clinic drafts a social post about a new treatment offering, and Lekha flags the phrase implying guaranteed pain relief as a health claim needing review, holding it until the treating physiotherapist approves the wording.
See the health and wellness playbookLekha drafts service-package descriptions, appointment-reminder copy, and promotional posts while keeping any wellness-outcome claim flagged for owner review.
A day spa asks Lekha to write copy for a new facial package, and Lekha drafts the description but flags a line implying a specific skin-improvement result as needing the owner's confirmation before it's used in an ad.
See the spas and salons playbookMore from Lekha
Everything else it handles
Learn and apply brand voice
The business gets copy that consistently sounds like them, in every language, without re-explaining tone in every prompt.
Learn moreGenerate copy for any channel
The business drafts a usable first version of an ad, caption, email, or product description in seconds instead of starting from a blank page.
Learn moreAdjust tone per piece
The business can dial a single draft from formal to playful (or vice versa) without rewriting it from scratch.
Learn moreWrite in Hindi and Hinglish
The business reaches Hindi-speaking and code-mixed audiences with copy that reads naturally, not like a literal translation.
Learn moreTranslate existing content
The business turns one piece of English copy into ready-to-use versions in other languages without hiring a translator for routine content.
Learn moreWrite seo-ready articles
The business gets a blog post or landing page draft built around a target keyword, ready to publish with minimal editing.
Learn moreQuestions
Straight answers.
Does this guarantee our marketing copy is legally compliant with ASCI or the Consumer Protection Act?
No — this is a detection and routing mechanism, not a legal-compliance guarantee. It flags language that reads as a regulated claim and holds it for human review, but whether a specific claim is actually substantiated and compliant is a judgment the designated reviewer has to make, ideally checking it against the business's own current, verifiable data, not something this classification step decides on its own.
What if it misses a genuine regulated claim and it gets published anyway?
Under-flagging a real regulated claim is treated as the single highest-severity failure mode across this entire product, because the consequence of a missed health or financial claim reaching customers is significantly worse than an unnecessary review delay. This is exactly why the detection needs ongoing tuning and monitoring rather than being treated as solved once it performs well on obvious, easy cases.
Won't flagging too much just mean reviewers stop actually reading what's flagged?
Yes, and this is a real, recognised risk on the other side of the same problem — if routine, safe language gets flagged constantly alongside genuine risks, reviewers can start rubber-stamping approvals without reading carefully, which defeats the purpose of the checkpoint just as thoroughly as under-flagging does. Both directions need active tuning over time, not a one-time calibration assumed to hold indefinitely.
Who actually has to be the reviewer for a flagged financial or health claim?
The business designates who that is — typically someone who can check the specific claim against real, current, verifiable information, like a compliance officer checking an interest rate against the approved rate sheet, or a licensed professional signing off on a health-related statement. The system enforces that a decision gets recorded by whoever is designated before publishing proceeds; it doesn't decide who that person should be.
The rest of your stack
It plugs into what you already run.
Lekha connects to your tools
No rip-and-replace — route risky claims to a human works alongside the systems already running your business.
- X
- MailChimp
- Webflow
- Shopify
- OpenAI
- Google Drive
- Figma
Coming soon