You wrote the testimonial. They just approved it.

A note from the archive: this post, from September 2026, follows the earlier KarmaLens method, which scored every reading on eight lenses from −5 to +5. Today a reading asks the decisive facts first and makes one call; the lenses below are a way of looking at the decision, not a score. See how a reading works.

The client replied in eleven minutes. "Looks great, ship it." You'd spent forty minutes on the paragraph. You know exactly which adjective was yours, because it's the one you were slightly proud of. And now it's on the homepage in their name, above their real logo, and it reads like something a delighted human typed at their own keyboard. Which it wasn't.

The customer said yes. You just can't stop noticing they said yes to your words.

The dilemma

This is a composite — the kind of situation we see often enough in the founder category to stitch into one clean example. No real person, real stakes.

Rohan runs a six-person B2B tool. His biggest account, a logistics ops lead named Priya, genuinely likes the product — she's renewed twice and mentioned in a call that it "saved her team a headache during peak." Rohan needs homepage social proof for a fundraise deck due Friday. He knows Priya is slammed. So he does the helpful thing: he drafts the testimonial for her.

He writes it warm and specific: a metric she's mentioned before ("cut our reconciliation time by roughly a third"), a line about the support team, a closing sentence with a punchy adjective Priya would never use — but that scans beautifully on a landing page. He sends it: "Wrote something up so you don't have to — edit anything that's off, or just reply 'good' and I'll run it."

Priya changes one word, softens "transformative" to "really useful," and replies "good, go for it." Legally, it's approved. She read it. She could have deleted it. She chose to sign.

Here's the detail Rohan hasn't said out loud: the reconciliation metric is real, but Priya never phrased it as a result of his product — she said it in passing, and he attached it. The endorsement is now on the homepage, in the deck, and quoted in an email to a second prospect. And every time Rohan sees it, some small part of him files it under "the numbers I built" rather than "the numbers I earned."

The read

Four of the eight lenses actually bite here. Two of them disagree, and the disagreement is the whole post.

Honesty: is the artifact true? (−2)

Honesty asks whether the words on the page match reality. Partly yes: Priya likes the product, the metric roughly exists, the renewal is real. But the sentence structure implies Priya sat down and chose to praise Rohan — and she didn't. The voice is manufactured. The adjective she'd never say is doing work she never endorsed. A reader assumes a testimonial is a customer's spontaneous account; this one is a founder's spec sheet wearing a customer's face. That gap is the falsehood, and it scores −2 — not a fabrication, but a staged authenticity, which is its own quiet lie.

Non-harm: who eats a cost here? (−1)

Non-harm looks for the person who pays. On the surface, nobody: Priya's happy, the prospect gets a true-enough claim, Rohan gets his proof. But Priya has now vouched, in public, for words she didn't compose and a framing she didn't quite mean. If the reconciliation number gets challenged by that second prospect, it's Priya's name attached to it, not Rohan's draft. He handed her a small reputational liability she never agreed to underwrite. Contained, reversible, but real. −1.

Wider welfare: what norm does this set? (−1)

Wider welfare asks what happens if this becomes the standard move. Ghost-written testimonials are already so common that buyers have learned to discount all of them — which means the founders who wait for a customer to write their own now compete against a page of fluent fakes and get punished for the honesty. Every staged quote taxes the currency of the next real one. −1.

Motive: what is the quote actually for? (+1, and this is where it fights honesty)

Motive is the interesting lens, because it partly defends Rohan. His intent isn't to deceive Priya — it's to reduce her effort. "Save the client time" is a genuinely considerate impulse; busy customers really do let good testimonials die in a drafts folder. There's a version of this where drafting a quote for approval is legitimate service. So motive scores a cautious +1 — not clean, but not predatory.

And that's the fault line. Motive says I was being helpful. Honesty says helpful to whom, exactly — her calendar, or your homepage? The tell is the adjective Rohan was proud of. You don't sneak a word you're proud of into someone else's mouth to save them time. You do it to make the artifact sing. The moment the draft optimized for the landing page instead of for Priya's actual sentence, motive stopped covering for it.

The verdict: run a testimonial you draft only if it survives Priya writing it herself — send her three of her own past phrases and a blank line, not a finished paragraph she can rubber-stamp. As it stands, this is consent to your words, not her endorsement, and the cost is that you'll never fully believe your own social proof again.

Watch what a committed call like that costs, and notice it's not the shape a general AI chatbot hands you. Ask a chatbot "I wrote the testimonial they approved — is that wrong?" and it will tend to find the reading that matches how you asked: lean toward permission and it reassures you that consent is consent; lean toward guilt and it validates the guilt. Sycophancy — models tilting toward the user's stated view — is a well-studied failure mode of conversational AI, not a one-off quirk. KarmaLens asks the decisive unknowns before committing to a call, uses the Gita's eight lenses as instruments on the dimensions the dilemma actually turns on — honesty, motive, non-harm, and so on — and makes one call that doesn't renegotiate itself to flatter your framing. Where a verse genuinely fits, it appears; it isn't bent into whatever you'd prefer it meant. That's the difference between a mirror that agrees with your posture and a reading that holds still while you move around it.

The takeaway: the read-it-aloud-in-their-voice test

You don't need us to catch a ventriloquized endorsement. Do this tonight with any testimonial you had a hand in:

1. Take the quote and read it out loud in the customer's actual voice — the way they text you, the way they talk on calls.
2. Underline every phrase that doesn't sound like them. The proud adjective. The clean metric they never quite said. The closing line that scans too well.
3. For each underline, finish this sentence: "This is here because it helps ___." If the blank fills with the customer (a real thing they meant, said better), it's editing. If it fills with the homepage, the deck, or the conversion rate, it's ghostwriting.
4. Then the harder question, the one this whole thing rests on: if the win were real, why did it need writing? A genuine result usually has a customer who'll say one true clumsy sentence. If you had to compose the enthusiasm, you may not have had the enthusiasm.

The fix is small: send them their own words back, not yours. "You mentioned reconciliation and the peak-week thing — would you be up for a line in your own words?" Slower. Less quotable. Yours to trust.

Staging one right now? Run it through a full eight-lens read and see where the honesty score lands before it hits your homepage. Which of your numbers would survive the customer telling the story unprompted?