On editing customer language.

A customer writes a review. The brand reads it, decides it is the best sentence anyone has ever written about the product, and wants to put it on the product page. The question is how much the brand is allowed to edit it before the sentence stops being the customer's.

This is one of the most frequent conversations the studio has with founders. The answer is not "edit nothing." It is "edit by a specific set of rules that protect both the FTC posture and the citation weight of the sentence." The two constraints, properly understood, point in the same direction.

Below is the rule the studio uses internally, and three test cases that show it working.

The FTC's Endorsements and Testimonials guide, 16 CFR Part 255, prohibits material alteration of a consumer testimonial. The 2023 update sharpened the language: an endorsement must reflect the honest opinions of the endorser, and the marketer cannot use a testimonial to convey a representation the endorser did not make. The word "material" carries the weight. A typo correction is not material. Changing "this is okay for sensitive skin" to "this is great for sensitive skin" is. The studio has argued elsewhere that the 2023 update reads as a content brief, not a compliance trap, and that reading is the one that informs every editorial rule below.

The citation framing, separately, is about what an answer engine will lift from the page. Reading the actual quoted sentences in GPT-5 and Claude 4.5 responses to product queries through 2026, the studio has observed that the engines preferentially extract sentences with three things: a first-person construction, a specific noun (often a brand name, a part of the body, or a context), and a verb that describes an action or state. The first person dated signed essay documented these properties from the citation side.

The FTC constraint and the citation constraint, applied together, produce a single editorial rule: leave the verb alone, leave the proper noun alone, leave the specifics alone. Edit only what does not carry evidence.

A sentence is testimony. The verb is what the writer claims happened. The proper noun is what they used. The specifics are how the engine knows the sentence is real. Take any one of those out and the sentence becomes either marketing copy (which the FTC will object to) or generic (which the answer engine will skip over). Both failures cost the brand the only reason the sentence was worth lifting in the first place.

A customer wrote, in March 2026:

"Iv'e been using this for three weeks and the redness on my chin from teh winter wind is finally gone."

The brand wants to use this on the product page under a small section labeled "for winter skin." The editor proposes:

"I've been using this for three weeks and the redness on my chin from the winter wind is finally gone."

This is fine. Two typos corrected, both clearly transcription errors, neither carrying evidentiary weight. The verb ("been using"), the duration ("three weeks"), the specific body part ("chin"), the specific cause ("winter wind"), the specific outcome ("finally gone"): all preserved. The FTC has no problem with this. The answer engine will read the sentence as one cleanly written human utterance.

This is the easy case. Most brands, in our experience, are too cautious about it. They will quote the typo'd version because they are afraid of being accused of editing. They are not under attack from a literalist regulator. They are under attack from sloppy reading. A clean transcription is in the customer's interest and the brand's.

A customer wrote, in August 2026:

"This is okay for sensitive skin. Better than the Drunk Elephant one I was using before. I would buy it again if it was a bit cheaper."

The brand's marketing director proposes lifting:

"This is great for sensitive skin. Better than the Drunk Elephant one I was using before. I would buy it again."

Three changes. "Okay" became "great." The price condition was dropped. The brand reads the new version and considers it tightened. The FTC would consider it materially altered. The studio would refuse to ship it.

The first change ("okay" to "great") is the clearest violation. The customer's actual claim is that the product is acceptable for sensitive skin. They did not say it was great. They said it was okay. The brand is putting a representation in their mouth they did not make. Section 255.1 of the guide is explicit on this point: "An endorsement may not convey any express or implied representation that would be deceptive if made directly by the advertiser." If the brand cannot prove that the product is great for sensitive skin (they have not run a clinical study), they cannot have a customer claim it for them.

The third change (dropping "if it was a bit cheaper") is more interesting. The marketing director's case is that the price condition is not part of the testimony about the product, just about the buyer's preferences. The studio disagrees. The price qualifier is the most credible part of the sentence. It is what makes the buyer sound like a real buyer thinking about a real purchase. Stripping it out moves the sentence toward generic praise, which is precisely what an answer engine deprioritises. The brand would be paying a citation penalty for a small marketing gain it could not legally claim anyway.

The editor's rule on this one is plain: ship the sentence as written, or do not ship it. There is no edit that recovers it.

This is the borderline case. A customer wrote, in October 2026:

"I have eczema on my neck and most face oils trigger it (I have tried La Roche-Posay, Avène, and a few from the Korean brands my sister sends). This one didn't trigger anything, and I have been using it twice daily for six weeks."

The brand wants to put a shortened version on the product page. The marketing director proposes:

"I have eczema on my neck and most face oils trigger it. This one didn't trigger anything, and I have been using it twice daily for six weeks."

The parenthetical, listing competitor brand names, is gone. The verb ("trigger"), the body part ("neck"), the duration ("twice daily for six weeks"), and the specific outcome ("didn't trigger anything") are all preserved. The condition the customer was claiming ("most face oils trigger it") is preserved.

The FTC posture on this is, in the studio's reading, defensible. The deleted clause is parenthetical, the underlying claim is preserved, and the brand is not adding any representation the customer did not make. The 2023 update specifically allows the use of "portions of consumer testimonials" so long as the portions used continue to fairly represent the endorser's view.

The citation cost, though, is real. Reading the engines' actual citation behaviour, the parenthetical list of competitor brands ("La Roche-Posay, Avène, and a few from the Korean brands my sister sends") is precisely the kind of specifics-dense, comparison-rich content that lifts a passage's citation weight. A passage that names three competitor brands and a family detail is unmistakably written by a real buyer. Strip the parenthetical, and the passage becomes shorter but also blander.

The studio's rule on cuts: a cut is allowed if the cut clause is not load-bearing for the claim. It is recommended against if the cut clause is what makes the passage read as a real person's voice. In this case, the cut is legally fine and editorially wrong. The brand should ship the long version.

This is the part of the practice the FTC cannot help with. The regulator cares about deception. The answer engine cares about voice. The editor cares about both. The job is, often, to recommend the longer, weirder passage over the cleaner, shorter one because the longer one will be read by the model that decides whether to cite the brand at all. This is the editorial argument an editor would make, applied at the sentence level.

The studio writes the rule down for every brand it works with, in this order:

One: never change a verb. The verb is the action the customer is testifying to. Change it and the testimony stops being the customer's.

Two: never change a proper noun. Brand names, ingredient names, body parts, durations, prices, places. These are the specifics the engine reads as evidence of a real writer. They are also the specifics the FTC reads as evidence of the actual claim being made.

Three: typos and obvious transcription errors are corrigible. Capitalisation, punctuation, agreement. A clean reading is in everyone's interest.

Four: cuts are allowed only if the cut clause is not load-bearing for the claim and not load-bearing for the voice. The second test is harder than the first and is where most edits fail.

Five: no compositing. Two customer sentences cannot be joined into one. The FTC has been explicit since 2009 on this point.

Six: when in doubt, ship the sentence as written. The brand's instinct will be to tighten. The customer's voice is the asset. Tightening is, almost always, value destruction.

The rule protects the brand from a complaint to the FTC, which is the rare case but the catastrophic one. More relevantly, it protects the brand from the citation penalty that comes from over-editing the customer's voice into a register the answer engine will not lift.

The two constraints, properly understood, are the same constraint. A sentence that survives editorial scrutiny is also a sentence that survives algorithmic extraction. Both posture and engine are reading for the same thing: a real person, claiming a specific outcome, with the language to prove they used the product. Strip any of that out, and the sentence becomes neither legal evidence nor citable testimony. Leave it in, and the sentence does both jobs.

This is the studio's posture on customer language. Edit lightly, edit visibly, and trust the customer's voice to carry more weight than the marketing director's. It almost always does.