Why can't Copilot find our sales content?

Copilot cannot find sales content for one of two reasons. The content sits outside what Copilot retrieves from, or it sits inside and nothing marks it as the one piece worth returning. Microsoft 365 Copilot grounds its answers in the Microsoft Graph and the semantic index, reaches only what the person asking already has permission to open, and ranks what it finds on how closely the text matches the question.

The people asking are enablement leads, marketing teams and sales operations who rolled Copilot out, watched it summarize a meeting well, and then watched it miss the pricing deck that has been in SharePoint for a year. This page covers how Copilot retrieves content, the seven reasons a sales deck does not come back, why the copy it does return is often the wrong one, and what a retrieval layer scoped to approved content adds beside it.

Why can't Copilot find our sales content?

Copilot cannot find sales content when the content is unreachable, unreadable or indistinguishable. Unreachable covers files that never entered Microsoft 365 and sites that have been taken out of search. Unreadable covers words that live inside a picture. Indistinguishable covers the common case, where the deck is indexed, the deck matches, and so do eleven other things.

The third case is the one that surprises people, because every part of the system is working. Retrieval found the material. Ranking put the closest text first. The assistant cited what it used. What the organization wanted was the approved deck, and approval is not something the index records, so it played no part in the answer.

Sales content is unusually exposed to this. A proposal deck is copied for each account, so one approved argument becomes forty near-identical documents that share almost all of their text. A general-purpose index sees forty candidates with no parent among them, and the one the compliance team would have chosen is not distinguishable from the thirty-nine built out of it.

How does Microsoft 365 Copilot retrieve content?

Microsoft 365 Copilot retrieves content by querying the Microsoft Graph and the semantic index, then passing what it finds to the language model as grounding for the prompt. Microsoft generates a tenant-level semantic index from text-based SharePoint Online files, and a user-level index covering the person's own mailbox and the documents they work in.

Four published properties of that pipeline explain most of what teams observe. Graph-grounded responses cover Word, PowerPoint, PDF, OneNote and aspx content along with Copilot connector data, and PDF, PPTX and DOCX files are supported up to 512 MB. Grounding honors the user identity access boundary, so retrieval reaches only content the person asking is authorized to open. New documents on SharePoint Online sites reachable by two or more users are indexed daily, while updates to an already indexed document are indexed immediately. And a site can be set not to appear in search results at all, which takes it out of Microsoft Search and the semantic index together, with no option to exclude one and keep the other.

Tenant-wide scoping has its own history worth knowing. Restricted SharePoint Search gave administrators an allowed list capped at 100 sites while they reviewed permissions. Microsoft blocked new enablement of it on 31 July 2026 and retires it on 31 January 2027, with Restricted Content Discovery as the per-site successor that keeps a named site out of Copilot and Microsoft 365 search. A team that met a strangely small search surface during a Copilot rollout was often meeting that allowed list.

What are the reasons Copilot does not return a sales deck?

The table below names 7 reasons a sales deck does not come back from Copilot, what is happening underneath each one, and what the person asking sees instead. The first five concern reach and the last two concern the shape of the answer.

The split across the table matters more than any single row. The first five rows describe content Copilot never saw, and each of them has an owner somewhere in IT. The last two describe what happens once retrieval is working perfectly, and no amount of tenant hygiene changes either of them, because both follow from what the index holds rather than from how well it was configured.

Why does Copilot return an old version of a sales deck?

Copilot returns an old version of a sales deck because the index has no concept of a current version. It holds documents, their text and their relationships, and it ranks them against the question. A deck from two years ago whose wording matches the question closely is a better candidate, by that measure, than this quarter's deck whose wording matches it loosely.

Duplication turns that from an edge case into the normal case. One approved deck spawns a copy per account, per region and per pitch. Each copy is a full document in its own right, each carries almost all of the original text, and each is indexed on equal terms with the original. The organization knows which one is the parent. Nothing in Microsoft Graph records that relationship, so the assistant cannot use it.

Currency has the same problem as approval. A superseded price, a withdrawn claim and a rebranded logo are all still text, and text is what the index counts. Ranking on text is correct behavior for a general-purpose assistant reading a general-purpose document store. It is also the one behavior a sales organization cannot afford, because the cost of the wrong answer is a number in front of a customer. The version-control side of that argument sits on the page about linking slides between presentations, and the content that never enters the store at all is covered in the piece on what reps build with ungoverned AI.

Can Copilot find a single slide?

Copilot finds the presentation rather than the slide. Listed below are 5 properties of Copilot retrieval that hold however well a tenant is organized, because each one follows from the design of the index rather than from its configuration.

• The citation is a file: answers reference documents, so a question about the integration architecture diagram returns the deck that carries it and the reader opens it to find slide 47.
• Relevance is measured in text: a deck built around a subject and a deck that mentions it once compete on the same terms, and neither carries a signal about how central the subject is to it.
• Permission is the boundary rather than approval: retrieval reaches what a person can open, which is a security property. It says nothing about whether anyone reviewed the content or still stands behind it.
• Duplicates are separate documents: forty copies of one deck are forty candidates. The index holds no parent among them and no count of how many are derivatives.
• The index describes what exists: Microsoft Graph holds what the tenant stores, so an approved library and a folder of abandoned drafts present themselves identically.

None of the five is a defect. They are the properties of an index built to answer questions about an organization's documents, and they are the reason the slide question needs a second index built on a different unit. What that index has to hold is covered on the page about searching across PowerPoint files, and the SharePoint half of it on the page about why a SharePoint search returns the whole deck.

How does SlideHub work alongside Microsoft 365 Copilot?

SlideHub holds approved slides, documents, images and text as separately governed items in a central library, and puts search over that library in a task pane inside PowerPoint, Word, Excel and Outlook. Each item carries its own text, tags, thumbnail, owner and approval state, so a result is a slide somebody published deliberately rather than a file that happened to match.

SlideHub runs alongside Microsoft 365 Copilot and changes nothing about how Copilot retrieves. It administers no part of SharePoint, it does not crawl the tenant, and it reaches no file that nobody has published into the library. What it adds is a second, smaller retrieval surface where the content has been reviewed, which is the distinction a general index cannot make. The wider question of what a library owes an assistant before the assistant should answer from it is on the AI-ready sales content library page.

Assistants reach that surface through the Model Context Protocol integration, which exposes the approved library to tools including Claude and ChatGPT. Retrieval there is scoped to the approved set rather than to everything a person can open. SlideHub's own semantic search and template-based generation work from approved layouts and company content rather than from open training data, and each AI capability carries an organization-level on and off control. The retrieval mechanics for people rather than agents sit on the slide search page.

More than 500 organizations and over 10,000 professionals use SlideHub each month, among them KPMG, Thyssenkrupp, Netcompany and Bech-Bruun, and SlideHub is rated 4.9/5 on G2. It is a Microsoft 365 certified application, is SOC 2 Type II and Cyber Essentials certified, and hosts customer data in the EU (AWS Ireland) under GDPR. Procurement detail sits on the security overview. Teams can compare plans on the pricing page or book a 30-minute walkthrough and bring the question Copilot keeps answering from the wrong deck.

Frequently asked questions about Copilot and sales content

Why can Copilot not find our sales decks?

Either the deck is outside what Copilot retrieves from, or it is inside and nothing marks it as the one to use. Copilot grounds answers in the Microsoft Graph and the semantic index, reaches only content the person asking can already open, and ranks on how well text matches the question. Approval and ownership are not ranking signals.

Does Microsoft 365 Copilot search SharePoint?

Copilot grounds its answers in the semantic index, which is generated from text-based SharePoint Online files along with the user mailbox. Supported file types include Word, PowerPoint, PDF and OneNote, with PDF, PPTX and DOCX supported up to 512 MB. A site set not to appear in search results is excluded from Microsoft Search and the semantic index together.

Why does Copilot cite an old version of a document?

Because every copy is a separate document to the index and they compete on the same terms. The approved deck from March and four tailored derivatives of it all match the question, and relevance is measured in text rather than in whether anybody still stands behind the content. Nothing in the index names one copy as the parent.

Can Copilot find a single PowerPoint slide?

Copilot answers with documents and references to them, so a question about one slide returns the presentation that holds it. A 90-slide deck arrives as a citation and the work of finding the slide stays with the reader. Slide-level results need an index whose item is the slide rather than the file.

What sales content is invisible to Copilot?

Four categories. Files that never reached Microsoft 365, such as material on a local drive or a legacy file share. Sites excluded from search results. Content the person asking has no permission to open. And words that live inside a picture on a slide, which are graphics rather than text no matter how plainly a reader sees them.