Between January and May of this year, a 40-person EdTech company watched its district cold-email program produce the strangest chart its sales team had ever seen. Open rates sat locked at 58 percent, month after month. Replies fell off a cliff — from 126 on the January send to 60 in April, on lists of nearly identical size.

Nothing about the copy had changed. Nothing about the list had changed. What changed was who — or what — was doing the reading.

The company sells a reading-intervention platform to mid-size districts. Call them Brightwork Learning; the details here are a composite of vendor experiences from this spring, and the platform mechanics and industry figures come from published 2026 data. The numbers are the kind any K-12 outbound team can check against its own dashboard.

The Problem: Opens Up, Replies Down

Brightwork’s first theory was subject-line fatigue. The team rotated in fresh subject lines for the March send. Replies kept falling. The second theory was send timing; they moved from Tuesday mornings to Wednesday. April was the worst month yet.

The diagnosis finally came from an unglamorous place: an MX-record lookup on the full contact list. Seventy-one percent of Brightwork’s district contacts sat on Google Workspace — and Gmail had spent early 2026 rolling out Gemini-powered AI Overviews that open messages by machine, compress them into a two-line summary, and present that digest to the reader first.

That one fact explained the whole chart. The 58 percent open rate was partly machines opening email to summarize it. The falling reply rate was humans reading two machine-written lines, finding nothing worth acting on, and moving on. Industry data backed this up: click-through rates across billions of tracked emails slid from roughly 4.35 to 3.93 percent after summaries launched, even as measured opens climbed.

Brightwork’s pitch was not surviving compression. Their standard first email opened with three sentences of warm-up — a seasonal greeting, an acknowledgment of how busy administrators are, a soft introduction of the company. The actual offer, with its strongest proof point, lived in paragraph three. Gemini’s summary never got there.

The Rewrite Rule

The fix Brightwork settled on fits in one sentence: the first two sentences must carry the entire pitch — the named role, the concrete outcome, and the ask — because two lines are all the summary keeps.

Everything the team knew about AI summarization pointed the same direction. The models keep proper nouns, numbers, named outcomes, and requests. They discard pleasantries, context-setting, humility padding, and anything that reads as filler. An email built like an inverted pyramid survives; an email built like a letter does not.

Brightwork rewrote its full 12-email outbound library against that rule — a library built on the same skeleton as the 5-email sequence that turns K-12 lookers into buyers, with each touch now front-loaded. Here is the before-and-after on email one.

Before

Subject: Quick question about literacy outcomes

“Hi Dana, I hope your summer is off to a great start! I know how hectic the weeks before school starts can be, so I’ll keep this brief. I wanted to reach out because we partner with districts like yours to support literacy growth in the elementary grades. Our platform, ReadBridge, uses adaptive assessment to…”

Gemini’s summary of that email: “Vendor introduces ReadBridge, a literacy platform, and requests a call.” Generic, forgettable, and indistinguishable from forty other vendor emails that week.

After

Subject: 3rd-grade reading scores up 11 points in 14 districts

“Hi Dana — ReadBridge helped 14 mid-size districts lift third-grade reading scores an average of 11 points in one school year, and the same intervention model fits what Riverside is doing with its elementary literacy block. The full results are one click away; if they look relevant, I’d value 15 minutes during your first week back in August.”

The summary of that version has no choice but to keep the number, the grade level, the district count, and the ask. The AI is now selling the meeting.

The Six-Week Test

Brightwork did not switch everything at once. From mid-April to the end of May, the team ran a split test across 5,400 Google-district contacts: 2,700 received the old library, 2,700 the rewritten one. Same roles, same cadence, same sender infrastructure.

The results, per arm:

• Replies: 30 on the old copy (1.1%) vs. 41 on the new copy (1.52%) — a 38 percent lift.
• Meetings booked: 9 vs. 12 — up a third.
• Unsubscribes and spam complaints: flat in both arms.

Worth pausing on: the list, the offer, and the sender were identical. The entire lift came from moving the pitch into the two sentences the AI actually keeps. At Brightwork’s average contract value of $18,000, three additional meetings a month at their historical 40 percent meeting-to-opportunity rate is real pipeline — roughly $21,600 in new monthly opportunity value from a copy edit.

What Failed, and Why

Two of the twelve rewrites underperformed the old versions, and both failures taught the team something.

1. The discount-first email. One rewrite led with a promotional offer: “20% off ReadBridge licenses signed before September 30.” Front-loading the discount made the summary read like a coupon, and Gmail routed more of those messages to the Promotions tab, where district staff rarely wander during the back-to-school crush. Outcomes belong in sentence one; discounts do not.

2. The stat-stuffed email. Another rewrite crammed three proof points into the opening sentence — score gains, time savings, and a satisfaction figure. The summaries were unpredictable: sometimes the AI picked the weakest number, sometimes it mangled two stats into one. The lesson became a house rule: one number per email, and make it the best one.

A third, quieter casualty was the curiosity-gap subject line. Teasers like “Quick question about literacy outcomes” lose their power when the digest underneath reveals the pitch anyway. Number-led subject lines matched the front-loaded copy and kept the message coherent from inbox row to summary to full read.

What This Means for Your K-12 Program

If your district campaigns show steady opens and sliding replies, assume the AI inbox until proven otherwise. The audit takes an afternoon: run an MX lookup on your list to see how many contacts sit on Google Workspace, paste your current first-touch email into Gemini and read the summary it produces, then rewrite until the summary sells the meeting on its own.

The front-loaded format carries a bonus Brightwork did not expect: it forwards well. A two-sentence pitch survives the hand-off when an administrative assistant passes it up the chain intact — the dynamic covered in winning over the school admin assistant.

One caution before any rewrite project: none of this works on a decayed list. Machine opens now mask list rot — a stale contact still “opens” email through the AI layer, so the metric that used to warn you says nothing. Cleaning and updating your data has to come first, and summer turnover season is exactly when district rosters churn hardest.

That is where the rewrite meets the list. Brightwork’s two sentences only produced meetings because they landed in front of verified people in the right roles — the curriculum directors and assistant superintendents who own the literacy budget. K12 Prospects maintains accurate, verified, role-segmented contact data for exactly that job: five million K-12 contacts, so the best two sentences you have ever written reach a real administrator who can say yes.

The AI inbox is not the end of K-12 cold email. It is a second reader with no patience — and the vendors who write for it first are quietly taking the meetings everyone else is still chasing.