File-Level Archiving Is GA — And It Quietly Pulls Files Out of Copilot

What changed in July

Microsoft 365 Archive can now archive individual files and folders instead of only whole sites. Public preview landed March 30, 2026; general availability rolled out from late June through late July 2026, enabled by default in tenants where Microsoft 365 Archive was already active.

The old whole-site model was close to useless for the situation almost every enterprise is actually in: an active site carrying years of inactive content. You could not archive the 2018 project folders without archiving the site that the 2026 project is running in. File-level granularity fixes that, and it is a genuine improvement.

It also introduces a tradeoff that most of the coverage is treating as a bonus feature. We will get to that.

---

The money, stated plainly

| Tier | Rate | Annualized |

|---|---|---|

| Base tenant pool | 1 TB + 10 GB per licensed user | included |

| Overage (standard) | $0.20 / GB / month | ~$2,400 / TB / year |

| Microsoft 365 Archive | $0.05 / GB / month | ~$600 / TB / year |

That is a 75% per-GB reduction on archived content. On 10 TB of inactive overage, roughly $24,000/year becomes roughly $6,000/year. Retrieval is metered separately, so your effective savings depend on reactivation behavior — a point we will come back to, because it is where naive archive programs lose their gains.

---

The reality check: archiving does not free quota

This is the single most expensive misunderstanding about the feature, and it catches experienced admins.

Archived content still counts against your storage footprint. You are not reclaiming space. You are paying a lower rate on the same bytes.

If you are sitting at a quota ceiling and expecting an archive pass to create headroom, it will not. The things that create headroom are different:

• Version history trimming — often the single largest recoverable footprint in a mature tenant
• Recycle bins — both stages, including the site collection second-stage bin
• Preservation Hold Library — retention policy residue that silently accumulates and that most admins have never looked at
• Actual deletion of content that has no retention obligation

Archiving lowers the bill. Cleanup creates room. They are different projects with different outcomes, and conflating them is how a storage program delivers a smaller invoice while the quota alarm keeps firing.

---

The part nobody is pricing in: Copilot goes blind on archived content

Archived files are excluded from search and from Copilot's index.

Nearly every write-up on this feature frames that as an upside — archiving removes stale documents, so Copilot stops surfacing the 2019 version of the expense policy. And sometimes that is exactly right. Noise reduction in Copilot grounding is a real problem and archiving is a real lever against it.

But framing it purely as a benefit is a mistake, because the same mechanism runs the other way:

If you archive content Copilot needed, answer quality degrades and nothing tells you.

There is no warning. No "this answer would have been better with archived content" flag. No degradation alert. Copilot simply answers from a smaller corpus and sounds exactly as confident as it did before. The failure is silent, and it surfaces weeks later as "Copilot has gotten worse lately" with no obvious cause and no changelog to point at.

Why "inactive" is the wrong test

Storage tooling ranks candidates by last-accessed date. That is the correct signal for cost. It is a poor signal for AI relevance, because the two properties are close to uncorrelated.

Consider what scores as inactive but is high-value to Copilot grounding:

• Signed contracts and MSAs. Nobody opens them for years. They are the authoritative answer to most commercial questions.
• Closed-project post-mortems and design decisions. Rarely read. Precisely the institutional memory Copilot is supposed to make retrievable.
• Superseded-but-referenced policy. The current version is what people open; the prior version is what explains why a control exists.
• Regulatory submissions and audit evidence. Touched once, then dormant, then urgently needed.
• Completed engagement documentation. The exact corpus an internal "how did we solve this before" prompt depends on.

Every one of those is a top-ranked archive candidate by last-accessed date, and several of them are the highest-value grounding content in the tenant.

Now the inverse — content that is genuinely safe to archive:

• Superseded drafts where a canonical final exists
• Build artifacts, exports, and generated reports
• Duplicated attachments and mail-drop residue
• Personal working copies in departed users' areas
• Media assets from completed campaigns

The distinction is not activity. It is whether the content is authoritative for a question someone might ask.

A test that actually works

Before archiving a content class, ask: *if someone asked Copilot a question this content answers, would we want the archived version in the answer?*

• Yes → do not archive, or archive only after confirming a canonical active copy exists elsewhere
• No → archive freely
• Unsure → sample ten files and read them. The answer becomes obvious fast, and ten minutes here is worth more than any automated scoring pass.

This reframes the whole exercise. Storage tooling is optimizing for one variable. You need to optimize for two, and only one of them shows up on the invoice.

---

Retrieval is a plan, not an afterthought

Reactivation timing varies by how long content has been archived, and retrieval carries its own cost. Treating archiving as a one-way door is how programs get into trouble.

Before any bulk pass, settle:

• Who can trigger reactivation, and does that require a ticket, an admin, or a self-service path
• Expected latency per retention tier, tested rather than assumed
• Which content is off-limits because delayed retrieval would break a legal hold, a regulatory response window, or an operational SLA
• What retrieval costs at your expected volume, and whether a chatty department could erase the savings

That last one is not hypothetical. An archive pass that saves $18,000 a year and then triggers constant reactivation because someone archived an actively-referenced library nets out to work you did for nothing.

Legal hold content deserves specific attention. If a matter is live or reasonably anticipated, archiving anything in scope introduces retrieval latency into a process that has deadlines set by someone other than you. Coordinate with legal before, not after.

---

A rollout that will not hurt

Step 1 — Find out what you are actually paying. Establish current overage in GB and dollars. If you are not over your allocation, archiving saves you nothing and this entire project is premature.

Step 2 — Clean before you archive. Version history, both recycle bins, Preservation Hold Library. This is where headroom comes from, and it is free. Archiving dirty data means paying to store duplicates at a discount.

Step 3 — Classify by grounding value, not activity. Run the Copilot test above against your largest inactive content classes. Produce two lists: safe-to-archive and grounding-critical.

Step 4 — Confirm the default-on state. If Microsoft 365 Archive was already active in your tenant, file-level archiving likely arrived during the June-to-July window without an opt-in. Verify who has access to it and whether anyone has already used it.

Step 5 — Pilot on one content class. Something unambiguously safe — completed campaign media, build artifacts. Measure the billing delta and test a retrieval end to end, including latency.

Step 6 — Write the retrieval runbook before widening. Who, how, how long, how much.

Step 7 — Widen by class, never by date sweep. A date-based bulk archive is the mechanism by which organizations accidentally remove their institutional memory from Copilot in a single afternoon.

---

Our read

File-level archiving is a good feature and the economics are real — 75% off inactive overage is not a rounding error at enterprise scale.

But it has quietly changed what an archiving decision *is*. It used to be a pure infrastructure call: cold storage is cheaper, move the cold things. Now every archive decision also decides what your AI can see, and the two objectives do not point the same direction. Cost optimization says archive by last-accessed date. Copilot quality says keep anything authoritative, however dormant.

Organizations that run this as a storage project, using storage tooling, optimizing a storage metric, will hit their savings target and degrade their Copilot deployment at the same time — and because the degradation is silent, they will not connect the two.

Run it as a content governance project that happens to save money. Same savings, and you keep the corpus your AI investment depends on.

---

Need help separating grounding-critical content from genuinely cold storage before an archive pass? See SharePoint Consulting and SharePoint & Copilot, or get in touch.

Sources

• File-level archiving comes to Microsoft 365 Archive (public preview) — Microsoft Community Hub
• Microsoft 365 Archive file-level archiving: SharePoint cold storage for individual files — 4sysops
• SharePoint file-level archiving limitations — Orchestry