In late 2026, the enterprise AI search market has shifted from retrieving answers to managing permissions, as key players have changed their strategies. Microsoft made Copilot Search free with eligible licenses, Salesforce restricted third-party indexing of Slack data, and Google removed public pricing details. Consequently, organizations must choose between platform-bundled access control and neutral, customer-hosted solutions like Glean or self-built Azure AI Search layers.
Every demo in this category is the same. Someone types "what did we decide about the Q3 pricing change," a paragraph appears, three citations sit underneath it. Glean, Microsoft 365 Copilot, Google Gemini Enterprise and Dust all do this in a sales call. The products diverge somewhere else, in a place most evaluation scorecards do not cover: what happens to your company's permissions on the way to the answer.
That is the honest summary of the enterprise AI search market in late 2026. The retrieval problem is largely commoditized at demo quality. The access-control decision is permanent, expensive to reverse, and different in kind across the vendors selling you a search box.
What changed in 2026
Three moves reset the board.
First, Microsoft made search free. Copilot Search, a full unified search experience with semantic ranking, is now included with an eligible Microsoft 365 Copilot license at no additional cost, reaching over 100 connectors in the Microsoft Catalog plus custom and ISV connectors. Indexing synced connector data carries no extra charge for tenants with Microsoft 365 licenses, which retires the per-seat index quota that used to make Graph connectors a budget line. The documentation page was last updated November 3, 2026.
Second, source systems started taking data back. In May 2025 Salesforce rewrote Slack's API terms to prohibit bulk export of Slack data and bar its use for LLM training. On March 3, 2026 Glean discontinued its use of the Slack Discovery APIs and moved to a Real-Time Search connector that queries Slack live and keeps message bodies in memory only, with nothing written into the index. For a product whose pitch is one ranked cross-referenced index of everything, losing persistent indexing of a company's chat history is not a footnote.
Third, pricing went dark. Google removed published Gemini Enterprise seat prices from its site on October 9, 2026; before that the list figures were $21 per seat per month for Business and from $30 for Standard and Plus (Sentra, InSearch). Glean, Coveo and Guru never published prices at all. What remains is third-party procurement data: Vendr reports a median Glean contract of $98,890 a year across 55 deals, ranging from $29,880 to $208,897.
Three permission models, not three search engines
Glean mirrors source ACLs into its own index. Its stated rule is that if a user cannot open a document in the source system, that document cannot be used as context, returned as a result, or surfaced as a citation (Glean security docs, via The Daily Brief). Each of its 275-plus connectors has to understand a foreign permission model, which is why connectors are the expensive part of the product. Glean also offers something none of the platform vendors do: a customer-hosted deployment that runs the tenant as a managed service inside your own AWS or GCP account, with full data residency guarantees. You own the account, not the operations.
Microsoft never leaves its own identity plane. Copilot surfaces only content the user already has at least view permission to, enforced by Microsoft Graph (privacy docs). Inside SharePoint, Exchange and Teams that is a stronger guarantee than mirroring, because it is the same authorization check the file always used. Outside Microsoft, the guarantee is only as good as the connector, and one documented exception matters to EU buyers: models provided by Anthropic as a subprocessor are currently excluded from the EU Data Boundary.
Dust replaces your permissions with its own. An admin chooses which Slack channels, Notion pages or Drive folders sync into a workspace "space," and space membership decides who can read the content (Dust access-control docs). Source-system ACLs do not follow the document in. That is defensible for a team building agents on a curated corpus, with 12 connectors at $24 per seat per month billed yearly. As a company-wide search layer it means one admin decision at sync time becomes the access boundary for content that used to have fifty owners.

The sticker prices, and the ones underneath them
The arithmetic case, per user per month, from vendor pages and comparison checks made in October 2026:
- Microsoft 365 Copilot: $30 ($31.50 monthly), on top of a qualifying plan. E3 itself rose to $39 on July 1, 2026, so the real number on an E3 tenant is $69. Copilot Business for tenants under 300 users is $21, with an $18 promotional annual rate through December 31, 2026. The new E7 bundle (E5, Copilot, Agent 365) lists at $99.
- Amazon Q Business: Lite $3, Pro $20, plus index capacity billed at $0.264 per unit-hour (AWS pricing). The only genuinely usage-shaped model of the mainstream group.
- Moveworks, owned by ServiceNow: $150 per employee per year on its AWS Marketplace listing.
- Glean, Coveo, Guru, Gemini Enterprise: quote only.
On the installed base, Microsoft reported over 30 million paid Copilot seats in FY26 Q4 results, still a minority of the Microsoft 365 population. Gartner's 2025 survey of 187 IT leaders found 40 percent piloting and only 5 percent of completed pilots moving to a larger deployment, with analyst Dan Wilson noting that plans assume "a clean data environment, which is rarely the case." Glean by comparison is a $300 million ARR business as of May 2026, up from $208 million at the end of 2025 per Sacra's estimate, priced at $7.2 billion in its June 2025 Series F (Tracxn) and reportedly closed a $768 million growth round in August 2026 without disclosing valuation.
Buying neither is also on the menu. Building on Azure AI Search runs from $74 a month for a pilot to roughly $491 to $736 for mid-size production, plus model usage and the build, priced by content volume instead of headcount. That is the right shape when one group of questions carries most of the value, or when the content must stay in your own cloud.
The squeeze on the neutral index
The independent vendor's position has a structural problem. Platform vendors bundle a good-enough search from above, and they are the systems that cut off third-party indexing from below. Salesforce owns Slack. Microsoft owns Teams, SharePoint and Exchange. Google owns Workspace. All three sell a competing assistant.
Glean CEO Arvind Jain told TechCrunch in February 2026 that he sees OpenAI, Anthropic and Google as partners rather than competition, on the theory that enterprises want a neutral layer under the interface. That may hold for model vendors. It is harder against the productivity suites, whose own assistants now index the same corpora. The 2000s enterprise search cycle rhymes: a specialist index across everything was worth real money until the platforms bundled a good-enough one.
Nobody is good at this yet
Before you benchmark vendors against each other, note what the best public research says about the underlying task. Salesforce AI Research's HERB benchmark (arXiv 2506.23139) tests deep search over a retrieval pool of 39,190 synthetic enterprise artifacts, documents, transcripts, Slack messages, GitHub content and URLs. Even the best-performing agentic RAG method scored 32.96 on average, and the authors identify retrieval, not reasoning, as the bottleneck: systems fail to gather all the evidence, then reason confidently over the incomplete set. Two caveats belong with the number: the corpus is synthetic, and the benchmark did not evaluate Glean, Copilot or Dust.
The transferable conclusion is that an enterprise assistant evaluation is a retrieval evaluation. Build a set of questions from your own corpus whose answers you already know, weight it toward the hard cases (a decision that changed twice, a number that differs across three systems, a question whose correct answer is "you are not allowed to see that"), and score recall of evidence rather than fluency of the paragraph. Every vendor passes the fluency test.
What to actually do
The decision criteria that predict regret are not connector counts.
If the organization runs on Microsoft 365, start with Copilot, because the search layer now comes with the seat you are probably already buying. Buy the neutral index when you can name ten high-value systems the platform cannot reach, or when the index has to live in your own cloud, which is currently Glean's unique offer in this group. Do not build a company-wide search layer on space-substituted permissions.
Three dated traps belong in the same budget cycle as the purchase. Restricted SharePoint Search, the guardrail many tenants switched on to roll out Copilot early, is being retired: new enablement blocked July 31, 2026, full retirement January 31, 2027, with no automatic migration. Permissions cleanup in SharePoint and shared drives is work you must do whichever product you choose, because any of these tools will answer from whatever an employee can already open. And every consumption meter in the category, Glean's FlexCredits beyond 100 thinking-mode queries per user per week, Dust's monthly credits that do not carry over, Copilot's pay-as-you-go agent usage, needs a priced cap in the contract before the demo looks attractive again.
The seat price is the number in the deck. The permission model is the number you live with.
Sources
- Microsoft Learn: Microsoft Copilot Search and connector prerequisites
- The Daily Brief (Rajesh Beri): Glean vs Copilot vs Dust: Buy the Permission Model
- QueryNow: Enterprise AI search pricing comparison for 2026
- Sacra: Glean revenue, valuation & funding
- Glean docs: product and feature retirements, Flex pricing, deployment options
- Computerworld: Salesforce changes Slack API terms to block bulk data access for LLMs
- TechCrunch: The enterprise AI land grab is on, Glean is building the layer beneath the interface
- arXiv: Benchmarking Deep Search over Heterogeneous Enterprise Data (HERB)
- Microsoft: FY26 Q4 press release, Copilot privacy
- Sentra: Glean vs Copilot, Gemini Enterprise pricing; InSearch: Gemini Enterprise pricing
- Vendr: Glean marketplace data; AWS: Amazon Q Business pricing
- TechPartner.news: Gartner: Microsoft Copilot hype offset by ROI and readiness realities
- SaaS Rise: Glean raises $768M for enterprise search expansion; Tracxn: Glean funding profile
- Dust: pricing, connections, access controls
Frequently asked questions
Is Microsoft Copilot Search still free in late 2026?
Yes, Copilot Search is included at no additional cost with an eligible Microsoft 365 Copilot license, covering over 100 connectors.
How does Glean handle user permissions compared to source systems?
Glean mirrors source ACLs into its own index, ensuring that if a user cannot open a document in the source system, it cannot be used as context or returned as a result.
What changes did Salesforce make to Slack API terms affecting AI search?
In May 2025, Salesforce rewrote Slack's API terms to prohibit bulk export of Slack data and bar its use for LLM training.
What were the previous seat prices for Google Gemini Enterprise?
Before removing published prices in October 2026, Google listed Gemini Enterprise at $21 per seat per month for Business and from $30 for Standard and Plus plans.
