Build vs. Rent: Why Owning Your Company's Knowledge Base Beats Renting SaaS Search

Build an internal knowledge base instead of renting one when the information your company depends on is unstable, expensive to reach, or already leaking out through tools nobody approved. That describes most companies right now. Renting still makes sense for commodity lookups. It stops making sense when the ground under the rental keeps shifting.

The web you built your workflows on is disappearing

For twenty years, "search the web" was a stable operation. You typed a query, got ranked results, and clicked through to a source. Engineers searched Stack Overflow. Support teams searched forums and documentation. Everyone, eventually, searched Google.

That stability is gone, and it went quietly.

In September 2025, Google disabled the &num=100 parameter, a URL trick that let tools pull 100 search results in a single request instead of the default 10.¹ No announcement. It just stopped working. An analysis of 319 properties found that 87.7% lost impressions in Google Search Console almost overnight, and 77.6% lost unique ranking terms entirely.² Google has never confirmed whether the change is permanent or said why it happened. The likeliest read: it wasn't aimed at SEO tools at all. It was aimed at cutting off ChatGPT, Perplexity, and Claude from scraping Google's own results to build better answers than Google's own AI Overviews.¹

At the same time, the platforms that held the web's actual knowledge are thinning out. Stack Overflow was fielding over 100,000 questions a month in early 2023. By December 2025, that number was down to roughly 3,600.³ Reddit, one of the most heavily cited sources in both Google's AI Overviews and standard search results, saw large swaths of its content de-ranked without warning after Google's 2026 core and spam updates.⁴ If any process at your company assumes stable, findable, third-party knowledge, that assumption is now wrong.

Three forces are breaking third-party knowledge retrieval

This isn't one trend. It's three, compounding.

• Google is closing the door on external access. Beyond removing num=100, Google's AI Overviews are quietly absorbing the clicks that used to go to source websites. Pew Research found that searches producing an AI summary led to a traditional result click only 8% of the time, versus 15% when no summary appeared, and just 1% of AI-summary visits resulted in a click on a link inside the summary itself. Users are also more likely to abandon the session entirely after seeing an AI Overview, 26% of the time versus 16% for ordinary pages. The web isn't just harder to scrape. It's being routed around.
• The communities that hosted real knowledge are shrinking or getting repositioned. Stack Overflow's collapse isn't purely about AI substitution. Strict moderation and bans on AI-generated answers accelerated the exodus of contributors who might otherwise have kept posting.³ Reddit's algorithmic demotion shows that even a platform AI systems rely on heavily can be repositioned overnight by an update nobody at your company controls.⁴ Wikipedia, YouTube, and Reddit remain the most frequently cited sources across both AI Overviews and standard search, which means a huge share of "AI-mediated" knowledge now funnels through a handful of platforms outside anyone's direct control.
• The rented knowledge tools have their own fragility. Enterprise search products promise to fix this by centralizing retrieval behind a subscription. They come with real costs and real seat math, which is the next problem.

Why is this a business continuity problem, not just an SEO one?

It's tempting to file all this under marketing's problem. It isn't. Knowledge workers already lose a huge chunk of the day just finding information. IDC pegs it at roughly 2.5 hours daily, about 30% of the workday, even when the retrieval layer is working normally. Now layer on the fact that the exact resources people used to search, from Google's top 100 results to Stack Overflow threads with a working answer, are drying up or getting harder to reach.²³

Support teams that resolve tickets by referencing community threads. Engineers who debug by searching for the exact error string. Onboarding docs that link out to third-party explainers. Every one of those workflows quietly assumed the public web would keep behaving the way it always had. Stack Overflow's decline from over 100,000 monthly questions to about 3,600 is the clearest warning sign available: a knowledge base an entire industry depended on for free, indefinitely, simply stopped refilling itself.³ There's no reason to assume the next community resource is exempt.

What does it actually cost to rent your knowledge layer?

The obvious counterargument is that buying enterprise search solves this. It doesn't solve it cheaply, and it doesn't solve it completely.

Glean, one of the category leaders, prices its base enterprise search license at roughly $45 to $50 per user per month, before add-ons.⁵ At even a few hundred seats, that's real annual money for a tool that indexes what you already own. And it sits inside a much bigger sprawl problem: the average enterprise now manages 305 SaaS applications and spends about $55.7 million a year on SaaS in total, while license utilization sits around 54%.⁶ Nearly half of what companies pay for, including knowledge and search tools, goes unused.

The governance side is worse than the cost side. 77% of IT leaders in one benchmark study discovered AI-powered features or applications running inside their organization that IT never approved, and 60% say they lack visibility into all the generative AI tools their own employees are using.⁶ Renting search from a growing list of vendors doesn't close that blind spot. It adds to it.

Shadow AI is already voting with its feet

Here's the part that should reframe the whole debate: employees have already decided the sanctioned tools don't serve them, and they've built workarounds without waiting for permission.

Gartner's research across 500 companies found that unauthorized AI tool use jumped from 41% in 2023 to 68% recently, with shadow AI tool usage up 156% over that period.⁷ People aren't doing this out of rebellion. They're doing it because the official retrieval tools are slow, incomplete, or don't cover what they actually need to find. The risk: roughly 54% of that shadow AI use has involved uploading sensitive company data, and about 76% of the tools involved fail to meet SOC 2 compliance standards.⁷ That's not a hypothetical exposure. That's company knowledge, and sometimes company data, already living outside anyone's control.

The honest read of that number isn't "crack down on shadow AI." It's "your employees have already told you, with their behavior, that they need an owned retrieval layer. Formalize it before it formalizes itself badly."

What changed to make owning search feasible now?

A few years ago, build your own knowledge base was a project for companies with a platform team to spare. Three things changed that math.

1. AI-assisted development collapsed build time. Productivity studies on AI-assisted engineering put the gain at 70 to 90 percent on routine tasks, the kind that used to make build-it-yourself a multi-quarter commitment.⁸
2. The population of people who can build got much bigger. Roughly 65% of users on some AI-assisted coding platforms have no traditional developer background at all.⁸ The person who understands what your support team actually needs to find can now build closer to the thing itself, instead of writing a requirements doc and waiting.
3. Managed primitives for the hard parts got better and cheaper. Auth, indexing, embeddings, and vector search are building blocks now, not bespoke infrastructure problems. That's the same shift already playing out in the build vs. buy math for internal tools more broadly.

CIO.com's framing of this cuts to the point: business users will build with or without IT's blessing, so the real job has moved from preventing builds to safely enabling them.⁸ That's the shadow AI pattern above, described from the other side of the org chart. Teams building their own retrieval and orchestration layers around that knowledge, sometimes with tools like Remy for structuring the internal workflows, are just doing openly what the shadow AI numbers show is already happening quietly.

How do you decide whether to build or rent internal search?

Not everything needs to be owned. The decision comes down to three questions.

1. Is this knowledge source stable and cheap to keep renting? Commodity, low-stakes lookups where third-party access is still reliable and the SaaS bill is modest are fine to leave rented. Don't rebuild what isn't broken.
2. Is this knowledge differentiating or operationally load-bearing? If your support team's ability to resolve tickets, or your engineers' ability to debug production, depends on a retrieval path you don't control, that's a build candidate regardless of what it costs to rent.
3. Is the rented version already fragile? If the source you depend on has already shown it can vanish or get repositioned overnight (num=100, Stack Overflow's decline, Reddit's de-ranking), treat that as your answer, not a warning to watch.

The point isn't to build everything in-house. It's to stop assuming that renting is automatically the safe, cheap default. On a knowledge layer this brittle, renting can be the riskier choice.

The bottom line

The public web stopped being a dependable retrieval layer sometime in the last two years, and most companies haven't updated their assumptions to match. Google restricted access to its own index, AI Overviews are keeping clicks inside Google instead of routing them out, and the community platforms that used to hold the answers are shrinking or getting silently repositioned. Renting a search SaaS on top of that doesn't fix the instability underneath it. It adds cost, seat sprawl, and its own governance blind spots.

Your employees have already answered the build-versus-rent question with their own behavior, by building workarounds nobody approved. The job now is to build the sanctioned version before the unsanctioned one becomes the only one that works. Treat your company's search and knowledge layer the way you'd treat any other piece of core infrastructure: something you own, maintain, and can trust to still be there next quarter, not something you're renting from a platform that might change the rules without telling you.