Paste the cold open, the warm follow-up, or the whole Outreach, Salesloft, or Apollo sequence. Get a 0-100 score in thirty seconds. The score predicts how a prospect will read the message: as a hand-written note worth a reply, or as the forty-seventh templated AI pitch in their inbox today. Sentence-level colour highlights show which lines are killing reply rates so SDRs, AEs, and founders can revise the generic phrases before the sequence goes live.
Cold B2B reply rates used to sit in the five to eight percent range for a well-targeted list and a half-decent message. Since late 2024 most teams running AI-drafted sequences report a typical drop to the one to two percent band. The cause is not algorithmic.
The opener that once signalled a human SDR spent five minutes on a LinkedIn profile now signals a generated sequence and the rest will not be worth reading. Senior buyers and procurement teams who see thirty cold emails a day pattern-match inside two seconds, archive the message, and never see the substance underneath. The substance might be genuine; the prose decides whether it gets read.
Modern spam systems do not run an AI classifier on inbound mail, but they do track per-domain spam-flag rates. When recipients hit Report Spam instead of replying, deliverability degrades. Generic AI outreach gets flagged at higher rates because the recipient has nothing concrete to attach to. The sending domain absorbs the cost of the templating shortcut over weeks.
A typical cold sales email runs four hundred to eight hundred characters, roughly sixty to one hundred and thirty words. In a ninety-word outreach, a single templated opener and a generic closer can account for forty percent of the entire message. Pattern density per word is roughly four times what it is in long-form content, which is why outreach needs to score higher than essays to clear the same bar.
SDRs running five or six step cadences face a second failure mode: even if step one lands, steps two through six often share the same AI register, and the recipient tunes out by the third touch. The sequence-level score matters more than any single email score because the cumulative pattern density across the cadence is what kills the reply rate over time.
The patterns that drop an Authenticity Score the most across cold outreach, warm follow-ups, and re-engagement sends. Fix the top three on every email and most sequences move from the mixed band into the reply-worthy band.
Universally dead. The single most recognised AI-flavoured opener in B2B. Most senior buyers archive on this line alone because it signals zero personalisation. Even hand-written outreach that uses this phrase reads as templated. Replace with a specific reference to the prospect or skip the pleasantry entirely.
Without specifics, this is the second-most flagged opener. Saying you noticed something on LinkedIn without naming what (a specific post, a recent role change, a product launch) signals a generated sequence. If you genuinely looked at the profile, quote the thing you actually saw.
Passive, vague, and dominant in ChatGPT's training distribution. The phrase delays the actual point of the message by one sentence and signals filler prose. Lead with the point itself: what you do, what you noticed, what would matter to this prospect.
Setup sentence, problem sentence, here is how we help sentence, three bullet points, ask sentence. This structure is the dominant output shape when you ask any LLM to write a cold email. Recognisable inside two seconds. Real human SDRs structure messages less predictably: sometimes two sentences, sometimes seven, occasionally a single question.
Save time, increase efficiency, streamline workflows, drive growth, unlock potential. Every prospect has read these phrases five hundred times this quarter. They are the linguistic equivalent of static. Replace with a specific outcome a comparable customer actually got, with a number attached.
Three flagged phrases stacked: would love to, hop on, and quick fifteen-minute call. This sentence single-handedly drops an Authenticity Score by eight to twelve points. The intent is fine. The phrasing is generated. Try: "If this lands, I can send the two-page version Tuesday" or "Worth twelve minutes this week?"
A closer that adds nothing. Real human emails just end. If a closer is needed, make it a concrete next step (I will follow up Thursday if I do not hear back) rather than a generic well-wish. Eight known variants appear in the standard LLM output distribution, so the scorer flags it consistently.
Five bands that map Authenticity Score to expected reply rate on cold and warm sends. The bands are calibrated specifically for short-form outreach where signal density is roughly four times what it is in long-form content.
Specific references, varied sentence lengths, no generic openers or closers, at least one concrete outcome or number. Buyers reading thirty cold emails a day stop on this kind of message. Reply rates typically observe the historical five to eight percent band when the rest of the targeting holds up. This is the target band for any cold first-touch in a high-value account.
One or two AI-flavoured phrases survived the edit, but the message has enough specifics to read as legitimate. Reply rates typically land in the three to five percent range. Worth one more pass on the flagged sentences before sending to a high-value account, but acceptable for second and third touches in a warm sequence.
The template structure is visible. Senior buyers and procurement teams who see high cold-email volume archive on the second sentence. Reply rates typically slide into the one to three percent range. Edit the red sentences before sending. If the message has a template opener and a generic closer both intact, fixing those two lines alone usually lifts the score twenty to thirty points.
Recognisable as generated within two to three seconds. Most recipients delete without finishing the second sentence. Reply rates collapse toward the one percent floor. Do not send at this score. The fix is structural, not wordsmithing: replace the opener, cut the bullet points, replace the call-to-action.
Raw or barely-edited LLM output. Recipients are more likely to flag as spam than reply. Sustained sends at this level can throttle deliverability for the sending domain over weeks. The fix is a complete rewrite, not an edit. Use the AI rewriter workflow or rewrite by hand from the prospect-specific angle outward.
Free covers twelve to twenty-five cold emails per day on the 10K daily detect budget. Active SDRs and AEs running larger volume usually run on Pro. RevOps teams running sequence-level audits across an SDR floor start on Business for the REST API. Full details on the pricing page.
Billed $89.88/year — Save $30
Billed $179.88/year — Save $60
Billed $359.88/year — Save $120
Yearly billing saves 25%. View full pricing →
A four-step loop that fits between draft and send. Most SDRs run it on every cold first-touch and on the lowest-scoring step of every active sequence. Total time under four minutes per email after the first run.
Start with the prospect-specific reason you are emailing. A recent funding round, a podcast they appeared on, a job posting on their team, a product launch, a public hire. Generate the body from that anchor, not from a generic template prompt. If the anchor does not exist, do not send.
Paste into TextSight. Inspect the sentence-level highlights. Take note of red sentences and the overall Authenticity Score. Anything under eighty on a cold first-touch needs revision before send; anything under seventy-five on a warm follow-up needs the same.
Red sentences move the score the most. Rewrite each one with a specific reference or a concrete number. If a sentence resists manual editing, run it through the AI rewriter in Light or Balanced mode. Skip Maximum for sales emails because aggressive rephrasing can soften pricing or deliverable claims.
Re-paste the revised version. Target eighty or higher for cold opens, seventy-five or higher for warm follow-ups. Send. Track reply rate against your historical baseline so you can verify the score is actually correlating with results on your list. Most teams see the correlation hold inside three weeks of testing.
Once a quarter, score every email in your existing Outreach, Salesloft, or Apollo sequence. The lowest-scoring step is usually a follow-up copied from a template library. Replacing that one step often lifts sequence-level reply rate measurably. Re-audit any sequence that has been running unchanged for more than six weeks.
If you change one thing about your outreach, change the opener. It carries thirty to forty percent of the perceived authenticity weight. Senior buyers decide whether to keep reading inside two seconds, and the opener is most of what they see.
Reference something the prospect did recently. A talk, a post, a hire, a product launch. Example: "Saw your Q3 deck on attribution modelling, the bit on email being undervalued matched what we are seeing across mid-market B2B." The specificity does the work; the opener earns the second sentence.
Name a shared connection, a shared customer, a shared event. Example: "Maria suggested I reach out after Tuesday's panel." A real mutual is the highest-trust opener available; a fake mutual destroys the conversation if discovered, so only use this when the connection is real.
Lead with the specific problem you suspect they are facing, based on observable signal. Example: "Your team is hiring three more SDRs, usually that means the current sequence is hitting reply-rate compression." The problem has to be one you can actually solve; otherwise the opener lands and the body deflates.
Lead with a concrete result a comparable customer got. Example: "We helped a ninety-person B2B platform recover twenty-two percent of attributed revenue inside a quarter after they cut email by forty percent." Numbers beat adjectives. A real outcome beats a generic claim every time.
Any of these four anchor types will move a sales email from the mid-fifties into the low eighties on the Authenticity Score, because the opener is the heaviest-weighted signal on short-form content. The body still needs work, but the opener is the highest-leverage single edit you can make. Most SDRs revise nothing but the opener and the closer on every email, and most of the reply-rate lift comes from those two edits.
The subject line decides whether the body matters. LLMs produce a narrow distribution that recipients pattern-match faster than any other signal. Subject-line scoring is the cheapest win available on any active sequence.
Quick question, Idea for [Company], Following up on [Topic], Helping [Company] with X. These have been used so heavily that open rates on the worst patterns sit below fifteen percent in most modern B2B funnels, against a healthy benchmark closer to thirty to forty-five percent. The distribution is small enough that one in three cold subjects is one of these four patterns.
The scorer treats the subject as a sentence and assigns a colour. Red subjects are recognised from the standard LLM distribution. Yellow subjects have one or two AI markers. Green subjects look like something a human typed for this one prospect. The subject and the opener combined account for sixty to seventy percent of whether the email gets read at all.
Specific to the recipient (a name, a project, a recent event). Short, four to seven words. They hint at the body without promising it explicitly. "Idea for Acme's email program" is fine. "Re: the Maria podcast Q3 attribution thread" is better. "Quick question" is dead on arrival. Test new subjects on small segments before scaling to a full sequence.
Each role has a different reply-rate failure mode and a different way the scorer fits into the day. SDRs ship volume; AEs ship targeted exec-level notes; founders ship investor and partnership outreach; freelance pitchers ship single-shot proposals.
High-volume outbound where any reply-rate lift compounds across hundreds of sends per week. Score every step in the active sequence each Monday and rewrite the lowest-scoring two. Track the sequence-level Authenticity Score average as a leading indicator of reply-rate health over the next three weeks.
Lower volume, higher value. Targeted exec-level outreach where one templated opener tanks the relationship before the first call. Score every bespoke message before sending to a target account. The scoring step adds two minutes; the cost of one templated opener to a CFO is much higher than two minutes.
Investor outreach, partnership outreach, candidate outreach. Generic AI-flavoured prose reads as "this founder did not bother." Score the message and rewrite anything that does not sound like the founder talking. Investor outreach in particular is where one templated phrase ends a relationship before it starts.
Cold pitching is the entire pipeline. Each message is high-stakes. Score every pitch before sending. A fifteen-minute revision can be the difference between a reply and silence on a five-figure project. The free tier handles this volume comfortably for most independents.
Quarterly sequence audits across an SDR floor. Score every step in every active sequence in one batch, surface the lowest-scoring steps across the org, and rewrite the bottom five percent. The REST API on Business automates this so the audit happens inside the existing sequence-management tool rather than in a separate dashboard.
The full rewrite workflow once the score flags red sentences in your outreach.
Open AI rewriter →Pre-send scanning for outbound sequences, proposals, decks, and RFPs across the SDR floor.
See the workflow →Sister page for job applications. The other short-form professional document where AI signal concentrates.
Read the guide →How the 0-100 score is calculated, what each signal weighs, and where the calibration comes from.
See the score →Free to try, no card. Sentence-level highlights, 0-100 score, subject line scoring, ten thousand detect characters daily.