Vertex AI Search Pricing vs Buying Enterprise Search

Most teams that price Vertex AI Search for internal use are asking a build or buy question. The engineering lead sees a per-query price measured in fractions of a cent and concludes that company search on Google Cloud will cost a few hundred dollars a month. Then the project meets the parts the query price does not cover: getting Slack, Jira, Confluence and SharePoint content into the index, keeping each employee's permissions intact, and building the screen people actually search from.

This piece puts numbers on every line, with Google's own figures read on October 1, 2026, works the bill out for a 500 person company, and compares it with the two ways to buy instead. We sell InSearch, one of those ways, so treat us as an interested party. Where building on Google is cheaper, the tables say so.

How much does Vertex AI Search cost?

On the pay-as-you-go model, Search Standard costs $1.50 per 1,000 queries and Search Enterprise $4.00 per 1,000, with core generated answers included on Enterprise. Advanced Generative Answers add $4.00 per 1,000 queries on either edition. Indexed data costs $5.00 per GiB a month. The first 10,000 queries and 10 GiB each month are free.

Google bills storage by the GiB-hour, $0.006849315, which comes to $5.00 over a 730 hour month. Savings plans bring that down to about $4.50 on a one year commitment and $4.00 on three years.

Is Vertex AI Search now called Agent Search?

Yes. Google's release notes record "Agent Search: Renamed from Vertex AI Search" on April 22, 2026, and the product now sits under the Gemini Enterprise Agent Platform. The console and the Discovery Engine API still use the older names, which is why both turn up in search results and invoices.

It is the fifth name in about two years. Google's own list of former names runs to Vertex AI Search, AI Applications, Agent Builder, Vertex AI Search and Conversation, Enterprise Search and Generative AI App Builder. The pricing page URL still carries the old generative-ai-app-builder path. None of this changes the price, but it matters when you compare a quote or a third-party article against Google's page: check that they describe the same product.

Can Vertex AI Search connect to Jira, Confluence, SharePoint or Salesforce?

Not anymore. Google's documentation, updated September 24, 2026, says "Connecting third-party data sources to Agent Search is no longer supported" and sends readers to Gemini Enterprise instead. Google Drive is the exception: Agent Search can still query it by federation, without copying the data into the index.

The Agent Search product page still says its connectors "ingest data with read-only access to key applications like Jira, Confluence, and Salesforce", so Google's marketing and its documentation currently disagree. The documentation is the one your engineers will hit. For an internal search project this changes the math completely. Every source other than Drive has to be exported, transformed and loaded into a data store by your own code, then kept in sync as pages are edited and permissions change. That pipeline is the expensive part of the project, and it does not appear on the pricing page.

What does Vertex AI Search cost for a 500 person company?

About $500 a month on Standard, $1,000 on Enterprise and up to $1,840 with advanced generated answers, for 500 employees running 20 searches each working day over 50 GiB of documents. That is Google's bill only, before any ingestion, interface or engineering cost.

The assumptions: 500 people times 20 searches times 21 working days is 210,000 queries a month, of which 10,000 are free, and 50 GiB indexed with 10 GiB free. The advanced answers add-on is billed on all 210,000 queries because the free allowance excludes it. Some articles still quote $400 to $600 a month for just 100,000 queries, or "Search with Advanced LLM" at $6 per 1,000; both are older price structures.

What does building employee search on Vertex AI Search really cost?

The query bill is the smallest line. The real cost is engineering: connector pipelines for every app except Drive, mapping each employee's identity and group membership so results respect permissions, an interface people will use, and someone on call when a sync breaks.

A realistic build needs at least one engineer for the first quarter and a meaningful share of one afterward. Pipelines for Slack, Jira and Confluence each have their own rate limits and API terms; Slack in particular restricts how third-party apps may store and index data from its API. Permission sync is where internal search projects usually stall, because a search tool that shows a salary spreadsheet to the wrong person gets switched off the same day. If your team does not already have that skill set, you will be recruiting a search engineer before you write the first pipeline, and that salary will outweigh years of query charges. Our page on permission-aware AI search lists what the permission layer has to do.

None of that makes building wrong. If you are building customer-facing search for a website or an app, Agent Search at $1.50 per 1,000 queries is a strong price, and you will not find a per-seat product that competes with it. The calculation only turns against building when the users are your own employees and the data lives in a dozen SaaS tools.

Is Gemini Enterprise cheaper than building on Vertex AI Search?

It is simpler, not cheaper on the bill. Gemini Enterprise Business starts at $21 per seat a month for up to 300 seats, and Standard and Plus start at $30 per seat. For 500 people on Standard that is about $15,000 a month, against about $1,000 for Agent Search queries, but the connectors, interface and permissions come built.

Google's own documents disagree on some limits. The product page says Business goes "up to 300 seats", while the editions documentation gives Business 1 to 500 users, and the Plus edition has no separate public price. Google also lists a pay-as-you-go option with no seat fee for organizations of 20 or more seats, still rolling out to a limited group of customers. We track those figures on Gemini Enterprise pricing. If you are comparing cloud search services, Amazon Kendra pricing runs the same exercise on AWS, where Kendra closed to new customers on July 30, 2026.

Should you build on Vertex AI Search or buy enterprise search?

Build on Agent Search when the searchers are your customers, the data is your own content and you have engineers to run it. Buy when the searchers are employees and the content lives in Slack, Jira, Confluence, SharePoint, Gmail and Salesforce, because that is exactly the part Google no longer connects for you.

Per month, building looks far cheaper, and for the Google line alone it is. The fair comparison adds an engineer's fully loaded cost to the build row, and at that point the gap closes within the first year for most mid-size companies. If you only need search for part of the company, the per-seat options scale down with you: InSearch starts at three seats, at $20, $32 or $49 per user a month billed annually, bought by card. Try it on your own documents with the search box at the top of this page before you commit an engineer to a pipeline project.