In the first quarter of 2026, something happened to email deliverability that no single monitoring dashboard could detect: two of the world's largest inbox providers — Google (Gmail) and Microsoft (Outlook/Microsoft 365) — simultaneously deployed AI-powered inbox prioritisation systems that fundamentally change how email is experienced by recipients. Gmail's Gemini AI Inbox rolled out in February 2026. Microsoft Copilot's "Prioritize My Inbox" feature reached general availability across Outlook clients in April 2026. Together, these two systems process email for over 3 billion combined users — and neither system works the way traditional spam filters did. This is not a spam filter update. This is a structural change in how email reaches human attention, and the email programmes that have not adapted are already losing ground they cannot see in standard delivery metrics.
The Simultaneous Rollout That Changed Everything
Email deliverability in the decade from 2014 to 2024 operated on a clear model: get to the inbox, and you've won. The spam filter was a binary gate — your email either passed into the inbox or was diverted to spam. ISPs made this decision based on authentication signals, IP reputation, domain reputation, and complaint rates. A programme that managed these factors well maintained stable inbox placement, and stable inbox placement translated directly to stable commercial performance.
The 2026 AI rollouts shattered this model. Gmail and Outlook have each deployed systems that go beyond the binary inbox/spam decision to create a ranked inbox experience — where delivered email is further sorted, prioritised, and in some cases automatically summarised before human eyes ever see it. The inbox is no longer a neutral container that delivers email for human decision-making. It is an active AI layer that interprets email relevance on behalf of each individual recipient, using that interpretation to determine how prominently (or how invisibly) the email is presented.
The commercial consequence is severe and, critically, invisible in traditional monitoring: a programme with 99% delivery rate and 95% inbox placement may have 40% of its technically-inboxed email effectively hidden by AI deprioritisation — never reaching the active reading view that generates clicks and conversions. The delivery metrics look fine. The campaign ROI quietly declines. Without monitoring effective inbox placement alongside traditional inbox placement, the problem cannot be detected from available data.
Gmail AI Inbox: How Gemini Filters and Ranks Email
Google's February 2026 announcement positioned Gmail's AI Inbox as a personal "proactive inbox assistant" built on Gemini 3. The user-facing features — AI Overviews that summarise email threads, AI Inbox that prioritises messages, AI-generated suggested replies — are the consumer experience layer on top of a deeper ranking system that affects all email in the inbox, commercial or personal.
The Gemini AI Inbox operates through several mechanisms simultaneously:
Semantic content evaluation: Gemini reads the full email content and evaluates it for relevance, clarity, and value density. This is fundamentally different from keyword-based spam filtering. Gemini is not looking for spam trigger words — it is evaluating whether the email communicates something useful and clear to the specific recipient. Overly promotional language, generic template copy, and vague openings score poorly on Gemini's relevance evaluation regardless of whether they trigger traditional spam filters. According to Folderly's deliverability research, which monitors millions of emails daily, Gmail's AI now explicitly evaluates "clarity, structure, and value density" as signals that influence inbox visibility.
Per-recipient personalisation: The critical difference from previous Gmail filtering is that Gemini's ranking decisions are personalised per recipient. Two recipients receiving the same email can have dramatically different inbox experiences — one sees it prominently, the other sees it deprioritised — based on each recipient's individual engagement history with the sender. This means aggregate inbox placement testing (sending to seed accounts) no longer fully represents the diverse inbox experiences of actual subscribers. A seed account with zero prior interaction with the sender sees a different placement outcome than an active subscriber with 12 months of click history.
AI Summary generation: For emails that Gemini processes, it generates a one-line inbox summary that often replaces the traditional preheader in the inbox preview. This summary is generated from the first 100-200 characters of readable email content — meaning the AI chooses what to show the recipient, not the sender. Emails with meaningful, specific, HTML-readable opening content get useful summaries. Emails where the first visible content is a hero image, a generic greeting, or promotional boilerplate get summaries that fail to communicate the email's value proposition. The AI summary becomes the first human-visible content — making the opening 200 characters of text more important than the subject line for a growing fraction of Gmail recipients.
Historical engagement as a continuous signal: Gemini continuously updates its per-recipient relevance model based on recent behaviour. A subscriber who clicked 8 of the last 12 sends has a strong positive engagement signal — Gemini's model predicts high relevance for the next send. A subscriber who received 12 sends without clicking has a degraded relevance signal — Gemini deprioritises the next send even if traditional authentication and reputation signals are perfect. This means engagement-based list management is no longer just about complaint rate protection — it is now a direct determinant of AI inbox visibility.
Microsoft Copilot "Prioritize My Inbox": The Outlook AI Filter
Microsoft's Copilot-powered inbox prioritisation reached general availability across Outlook clients — Outlook for Windows, Outlook for Mac, Outlook on the Web (OWA), and Outlook mobile — in April 2026. Unlike Gmail's Gemini AI Inbox, which is automatic and invisible to recipients, Copilot's "Prioritize My Inbox" is an opt-in feature that users can enable and customise. However, the Microsoft 365 enterprise adoption trajectory and the significant productivity gains users report when the feature is enabled suggest that adoption will be widespread among Microsoft 365 enterprise users within 12-18 months.
The Copilot prioritisation system operates on a different architecture from Gmail's Gemini, but produces similar commercial effects for commercial email:
How Copilot prioritises: The system analyses sender relationships (frequency of prior interaction, job title of the sender, whether the sender is in the recipient's contacts), message content and subject matter, timing relative to the recipient's work context, and the recipient's historical response patterns to similar email. It scores each incoming message as High, Normal, or Low priority. High priority messages are surfaced prominently. Low priority messages are technically delivered but visually demoted — easy to miss in the inbox view. Unlike Gmail's Gemini which integrates the ranking invisibly, Copilot's prioritisation is visible as labels on messages.
Commercial email in the Copilot ranking: Commercial email from unfamiliar senders — which includes virtually all marketing email from brands the recipient does not regularly engage with — scores Low priority under Copilot's model. The sender-relationship signal alone is enough to demote most marketing email, regardless of content quality. A B2B sender emailing a Microsoft 365 enterprise recipient who has never interacted with their email will have Copilot label the message as Low priority — even if the email is authenticated, relevant, and well-written. The relationship signal dominates.
The learning dimension: Copilot's prioritisation model learns from user behaviour over time. If a recipient consistently moves a sender's Low priority labels to the Focused view or clicks through to the sender's content, Copilot updates the priority model — eventually elevating future messages from that sender. This means first-contact engagement is more important than ever: the first email from a new sender establishes the relationship signal that Copilot uses for all subsequent email from that sender. A first email that earns a click is an investment in future inbox visibility.
The agentic future: Microsoft's 2026 roadmap includes "agentic inbox automation" — where Copilot can not only prioritise but autonomously respond to certain email types on behalf of the user, without requiring explicit approval for routine messages. When this feature rolls out broadly (expected in phases through 2026-2027), the definition of "email engagement" will need to expand to include AI-agent responses as a signal, distinct from human responses. Marketers will need to understand whether their engagement metrics include AI-generated replies — and what that means for the engagement signals ISPs and AI filtering systems evaluate.
The Double AI Problem for B2B Senders
B2B email senders face disproportionate exposure to the double AI problem because their primary audiences — business professionals at enterprise companies — are concentrated on exactly the two platforms deploying AI prioritisation. Gmail (Google Workspace) and Microsoft 365 together dominate enterprise email. A B2B sender targeting VP-level prospects at enterprise companies is sending to an audience that is 80%+ on Gmail Workspace or Microsoft 365 — both of which now have AI prioritisation deployed or rolling out.
The B2B-specific dynamics: (1) B2B contacts at enterprise companies have high Microsoft 365 penetration — Copilot's "Prioritize My Inbox" will reach the majority of enterprise business email users as adoption grows. (2) B2B cold email, where sender relationship is zero, gets Copilot's Low priority label by default — the sender has no prior engagement history to elevate the relationship signal. (3) B2B marketing email that goes to both Gmail Workspace and Microsoft 365 contacts faces two separate AI deprioritisation systems with different signals — requiring optimisation for both. (4) The traditional B2B engagement metric of "opens" is now doubly unreliable — Apple MPP inflates it from one direction, and Gmail Gemini auto-generating AI summaries without requiring a human open inflates it from another direction. Click rate is the only reliable engagement signal remaining for B2B programmes.
The Microsoft 365 cold email problem specifically: cold email to enterprise prospects who are Microsoft 365 users lands in an inbox where Copilot has identified it as Low priority because the sender-relationship signal is zero. The prospect sees the email labelled as low priority, with competing High priority messages visible at the top of their inbox. The cold email that previously competed on subject line and opening line quality now must overcome a visual hierarchy disadvantage before the recipient's eyes even reach the subject line. This structural disadvantage increases with every week of Copilot adoption among enterprise users.
"Effective Inbox Placement" — The New Metric That Matters
Folderly's deliverability researchers introduced the concept of "effective inbox placement" to capture the gap between traditional inbox placement (email reached the inbox) and actual inbox visibility (email was seen and could be acted on). Their data suggests that up to 40% of emails technically reaching Gmail inboxes are being deprioritised by AI filtering — delivered in the technical sense, but invisible in the practical sense.
The distinction is commercially critical. A programme measuring 95% inbox placement (traditional metric — 95% of delivered emails reached the inbox rather than spam) may have only 57% effective inbox placement (the 95% inbox fraction minus the 40% AI-deprioritised fraction = 57%). The revenue impact of 38 percentage points of lost effective visibility — at the same delivery metrics that look healthy in every dashboard — is significant and growing as AI adoption increases.
Traditional inbox placement testing cannot measure effective inbox placement because seed accounts, by definition, have no prior engagement history with the sender. AI prioritisation systems that weight engagement history cannot apply that weighting to a seed account that has never received email from the sender before. The seed account sees a "cold contact" placement outcome, which may be more pessimistic than the placement outcome for an actively engaged subscriber — or it may miss the AI deprioritisation problem entirely if the seed account's inbox client handles AI prioritisation differently from the bulk of the actual subscriber base.
How to approximate effective inbox placement without purpose-built tools: compare click rate trends against historical performance controlling for content and seasonality changes. If click rate is declining while traditional delivery metrics (inbox placement, authentication pass rate, complaint rate) are stable, AI deprioritisation is the most likely culprit. Segment the click rate decline analysis by email client (Klaviyo, Mailchimp, and most major ESPs provide email client attribution) — a click rate decline concentrated in Gmail while Outlook and Apple Mail click rates are stable specifically points to Gemini AI deprioritisation.
The Metrics Collapse: CTR Down 10%, Open Rate Now Meaningless
The data following Gemini's February 2026 rollout tells a clear story. Folderly's analysis across millions of daily Gmail messages shows click-through rates dropped from 4.35% to 3.93% — a 9.7% decline — in the weeks following the Gmail AI Inbox rollout. Simultaneously, reported open rates increased (because Gemini auto-opens emails to generate AI summaries, registering as opens in email analytics). The pattern: higher reported opens, lower actual clicks — the signature of AI-mediated inbox processing where more email is being "processed" by the AI but less is being actively engaged with by humans.
This metrics profile — rising open rates, falling click rates — is the opposite of what a programme should celebrate. It indicates that the AI is increasingly handling the "read" function while humans are clicking less. For any programme still using open rate as a primary engagement signal, this data is actively misleading: the rising open rate looks like improving engagement while the falling click rate reveals declining commercial performance.
The open rate is now a three-way ambiguous signal: (1) Apple MPP pre-fetching images inflates opens from Apple Mail users. (2) Gmail Gemini AI auto-opening emails to generate summaries inflates opens from Gmail users. (3) Actual human-opened email generates opens that look identical to the machine-generated opens in analytics. In 2026, reported open rate cannot be reliably decomposed into human opens and machine opens without client-level attribution and AI-open identification features that most programmes have not yet implemented.
Content Quality as a Deliverability Factor — What This Means Operationally
The single most consequential change that Gmail Gemini AI introduces to email deliverability is the elevation of content quality from a user experience concern to a deliverability signal. In 2025 and before, "good email content" was good marketing — it improved engagement, drove conversions, and built relationships. Poor content quality was a marketing problem, not a deliverability problem. Deliverability was determined by authentication, IP reputation, and complaint rates — all technical signals that operated independently of content quality.
In 2026, this separation has collapsed. Gemini's evaluation of email content — its clarity, structure, specificity, and value density — directly influences where the email appears in the recipient's inbox. Content that Gemini evaluates as high-clarity and high-value earns prominent placement. Content that Gemini evaluates as generic, vague, or promotional earns deprioritised placement. The email copywriter is now, in a very real sense, contributing to deliverability outcomes — not through spam filter keyword avoidance (which is and always was a minor factor for reputable senders) but through content quality signals that the AI evaluates on behalf of every individual recipient.
The operational implications:
Subject line + first sentence coherence: Gemini evaluates whether the subject line and the first sentence of the email body tell the same story. When they are coherent — the subject promises something specific, the first sentence delivers on that promise — the AI summary reinforces the email's value proposition. When they are incoherent — the subject creates curiosity, the first sentence is a generic greeting — the AI summary may surface the incoherent opening rather than the email's actual value, undermining the open. This is the most immediate, actionable change email teams can make: ensure the first sentence of every email directly delivers on the subject line's promise.
Images with no text alternative: Gemini cannot read text embedded in images. An email where the key message exists only as text baked into a hero image has, from Gemini's perspective, no readable content. The AI summary generated from such an email is empty or misleading. The rule is categorical: every meaningful message in a commercial email must also exist as readable HTML text, not only as image-embedded text. "One big image" email design is no longer defensible from either an accessibility or a deliverability perspective.
Semantic HTML structure: Gemini processes email HTML the way a sophisticated screen reader does — following the semantic structure of heading elements, paragraph elements, and list elements to understand the email's information hierarchy. Emails with proper semantic HTML structure (h1/h2 headings, paragraph text, bulleted lists as li elements) are more legible to the AI and generate more useful summaries. Emails built entirely from div elements or table cells without semantic markup are harder for the AI to parse.
Front-loading value: The first 100-200 characters of readable text are disproportionately important because they drive the AI summary generation. The standard email template structure — logo at top, hero image, then text — means the first readable content Gemini encounters is often generic boilerplate text that appears below the hero image. Restructuring templates to include a meaningful value-communicating sentence in the first visible text position (before or alongside the hero image) dramatically improves the quality of AI-generated summaries for Gmail recipients.
The Adaptation Playbook: What to Change Now
The AI filter adaptation playbook for commercial email programmes in 2026 operates on four levels: metrics, content, list quality, and relationship signals.
Level 1 — Fix the metrics: Stop optimising for open rate. Build primary reporting around click rate per delivered (CTD), revenue per delivered, and Gmail Postmaster Tools spam rate. Add email client attribution to click rate analysis — track click rate separately for Gmail, Microsoft, Apple Mail, and other clients to identify AI deprioritisation patterns by client. A click rate decline concentrated in Gmail while other clients are stable is the signature of Gemini AI deprioritisation. A click rate decline concentrated in Microsoft is the signature of Copilot deprioritisation.
Level 2 — Upgrade content for AI readability: (1) First sentence delivers on subject line promise — no exceptions. (2) Every meaningful message exists as HTML text, not only as image-embedded text. (3) Use semantic HTML structure — headings, paragraphs, lists — not just visual formatting. (4) Pre-header and opening sentence are redundant — they communicate the same value, not two different things. The subject line teases; the first sentence delivers; the email body expands.
Level 3 — Intensify engagement-based list management: Gemini's AI ranking directly weights engagement history. Disengaged subscribers who have not clicked in 60+ days have weak positive engagement signals in Gemini's model — their next email from you is deprioritised before anyone reads the subject line. Engagement-based suppression is no longer just a complaint rate protection measure — it is a direct inbox visibility optimisation. Suppress aggressively; the active list that remains will earn better AI ranking for every future send.
Level 4 — Build relationship signals deliberately: For both Gmail (Gemini) and Outlook (Copilot), prior positive engagement history is the highest-value inbox placement signal. The welcome sequence — the first 5-7 emails after a subscriber joins — is now the most important inbox placement investment in the entire programme lifecycle. A subscriber who clicks through the welcome sequence has established a positive engagement history that signals relevance to both AI systems for all future sends. A subscriber who ignores the welcome sequence starts with a weak engagement signal that will compound into progressively worse AI inbox placement with each subsequent send. Invest disproportionately in welcome sequence quality; the inbox placement dividend is permanent.
The email programmes that adapt to the AI filter era — understanding that content quality, first-sentence coherence, HTML semantics, and engagement-based management are now deliverability variables as much as marketing variables — will be the ones that maintain effective inbox placement as AI adoption increases. The programmes that continue optimising for pre-2026 metrics (open rate) while ignoring the AI-mediated inbox experience will see their commercial performance continue the quiet decline that the Gemini and Copilot rollouts already initiated for the unadapted majority. The disruption is real, it is measurable, and the adaptation path is clear.