First-Hand Versus Research-Only Product Reviews: The ABS Editorial Bar

What “first-hand” means on ABS

A first-hand review answers the question “did the operator use this thing, and what did they learn.” The body speaks in the operator’s voice about their own context: the apartment, the mac mini M4, the Hermes gateway in front, the cron-rig, the 7-day rolling traffic on this VPS. The product is the lens, not the story.

A research-only review answers the question “what does the rest of the internet say about this thing.” The body paraphrases Amazon’s bullet points, threads Reddit comments together, summarizes a vendor product page. Without a human-test signal, this is content marketing. ABS does not publish content marketing.

Why this is the line

Two reasons.

1. Trust. A reader knows the difference. A first-hand review is anchored to an authentic context; a research-only review is interchangeable with every other unboxing site. Trust compounds: each first-hand review slightly raises the trust bar of the whole site, so subsequent reviews are read more attentively.

2. SEO. Google has spent 8 years (from Panda through Helpful Content Update 2024) deprecating research-only content. The Helpful Content signal specifically rewards “was this written by someone who has used the product” and specifically punishes “is this what everyone else is writing.” A first-hand review keeps ABS in the rewarded bucket; research-only doesn’t.

The operator’s signal

The voice note must include at least three of:

1. Where you used it (the room, the surface, the device, the operating context)
2. What you used it for (the actual problem you were solving)
3. What you used before (the comparison product or workaround)
4. What broke or surprised you (a specific moment that mattered)
5. What you’d change about it (your list of gripes)

If the voice note lands and the botsitter-review skill sees fewer than three of these, the skill pushes back: “this sounds like a research summary; can you record again with what you actually did with it?” The operator re-records once and the post moves forward.

These five are not arbitrary. Each maps to a class of evidence that an LLM cannot fabricate without an obvious seam. Where you used it anchors the post in physical or operational reality — the operator’s actual desk, device, or workflow — and a fabricated room shows up the first time a reader tries the product in a different one. What you used it for captures intent, and intent is the thing Amazon bullet points strip out in favor of feature lists. What you used before forces a comparison that is falsifiable by anyone who owned both. What broke or surprised you is the single most diagnostic signal — surprises are hard to invent because they require the operator to know what should have happened. What you’d change about it converts passive experience into active critique, which is the difference between a review and a press release. When three of these five show up, the post is anchored. When fewer than three show up, the post is research-only and gets held.

The research-only trap

LLMs are good at producing research-only content — paraphrase the page, link the bullet points, summarize the threads. The botsitter-review skill is not supposed to do that. The system prompt for the skill explicitly forbids it:

“Write the body from the operator’s perspective — what they actually did with the product, what broke, what surprised them. Do not paraphrase the product page or vendor marketing copy. If the operator’s voice note does not contain first-hand signals, push back and ask for a re-record.”

This is one of those cases where the system prompt language matters more than the model choice. A model with the right prompt writes first-hand reviews; the same model with the wrong prompt writes research-only paragraphs. The skill’s prompt is the line.

The friendly-fail mode

Sometimes the operator sends a voice note on a product they’ve researched but not used. The skill’s response:

“This is great research, but the bar at ABS is first-hand only. If you want me to draft a piece on Amazon’s positioning of this product, that lives elsewhere. To ship here, you’d need to actually use the thing. Want to set a reminder to test it for two weeks, then re-record?”

That gives the operator a graceful next step instead of having to figure out where to file the work.

What not to do

• Don’t draft research-only content and ship it as a review — the editorial bar will catch up. The next promotion pass audits the prior batch’s first-hand-ness and removes anything that wasn’t.
• Don’t relax the bar for friends-of-the-site or high-profile products. The bar is a property of the site; it doesn’t bend.
• Don’t publish on the same product twice unless the second one is genuinely different (different operator, different context, different conclusion).
• Don’t fast-track by stripping the editor and shipping raw LLM output. The skill is the editor.

Verification checklist

#	Signal in voice note	Present in the post?
1	Where you used it	Yes / No
2	What you used it for	Yes / No
3	What you used before	Yes / No
4	What broke or surprised you	Yes / No
5	What you’d change	Yes / No

A post passes the editorial bar if 3 or more of these are present in both the voice note and the post. Less than 3 should pause the ship.

Sources

• ABS companion: Botsitter-Review Affiliate Link Convention
• ABS companion: Voice Note to Published Review
• ABS companion: Publishing a Product Review End-to-End

Last verified: 2026-07-14 against the live botsitter-review skill system prompt and 31 first-hand reviews published on ABS.