IDC Says AI Agents Guide 80% of Tech Buyers, Pressuring Vendor-Led Sales

IDC says 80% of B2B technology buyers now use AI agents to assist or perform buying tasks. That shift puts Google News, vendor websites, and sales teams into a new contest for machine-mediated trust.

The finding comes from IDC’s 2025 Tech Buyer Behavior Survey. It does not mean autonomous software has taken control of most enterprise purchasing budgets. It means AI has joined the buying group, helping people discover vendors, compare products, summarize evidence, and prepare decisions.

That distinction matters because enterprise buyers no longer need to visit every website that informs a purchase. An agent can collect information, reconcile product claims, and eliminate candidates before a salesperson sees an opportunity. The immediate conflict is not humans versus machines. It is vendor-controlled persuasion versus buyer-controlled evaluation.

IDC’s research also shows that digital channels remain central to complex purchases. However, separate Gartner and Forrester findings indicate that buyers still seek human validation when risk rises. AI can compress research, but trust still determines whether a recommendation survives procurement, security review, and executive scrutiny.

The 80% Figure Marks a Change in Who Controls Discovery

AI agents are becoming an active layer between technology vendors and the people evaluating them.

IDC’s buyer alignment blueprint states that 80% of buyers use AI agents to assist or perform purchasing tasks. The document draws on IDC’s 2025 Tech Buyer Behavior Survey.

The same blueprint says 71% of B2B technology buyers prefer digital channels, including for complex purchases. IDC describes the digital-first buyer as permanent rather than a temporary response to remote work or limited access to sales representatives.

These two findings describe related changes. Digital purchasing moved research away from scheduled sales interactions. AI agents now reorganize the information available within those digital channels.

An AI buying agent is software that researches, compares, or acts on purchasing requirements for a user. In a basic workflow, it may summarize product pages and customer reviews. A more capable system might compare configurations, draft a request for information, or check whether a vendor satisfies predefined requirements.

The agent does not need formal authority to influence a purchase. A general-purpose assistant can change a shortlist by deciding which sources to retrieve, which differences to emphasize, and which claims deserve skepticism.

That makes visibility inside AI-generated answers commercially important. A vendor can rank well in traditional search and still disappear from an agent-generated comparison. Its documentation may be difficult to retrieve, inconsistent across pages, or unsupported by independent evidence.

Google News remains relevant because current reporting can supply signals about acquisitions, product failures, leadership changes, security incidents, and customer disputes. Yet an agent may extract those signals without sending the buyer to the original result page.

This creates a different discovery funnel. The buyer receives an answer, while the agent decides which evidence enters that answer. Publishers and vendors can contribute information without receiving a visit, impression, or attributable lead.

IDC’s separate analysis of AI-mediated buying adds another constraint. Its 2025 survey found that 81% of buyers prioritize ethical AI use when selecting technology partners. That figure rises to 88% among C-level executives.

Therefore, getting mentioned is only one part of the problem. A vendor must also survive scrutiny involving responsible AI, data governance, security, and credibility.

The change is structural because agents can apply these filters before direct engagement. Traditional sales teams often discover objections during a call. An AI-mediated process can reject the vendor earlier, leaving the sales organization with no opportunity to answer.

The 80% figure should still be read carefully. “Assist or perform” covers a wide range of behavior, from summarizing a document to completing a defined task. IDC’s public blueprint does not divide the figure by autonomy level, purchase value, or transaction completion.

Even with that limitation, the direction is clear. AI has moved from a vendor feature into the buyer’s research process, changing how enterprise technology reaches a shortlist.

Google News Is Becoming Evidence for Agents, Not Just a Destination

Google News and similar channels increasingly serve as evidence layers that AI systems process on a buyer’s behalf.

Traditional search asks publishers and vendors to compete for a click. AI-assisted research asks them to compete for inclusion in a synthesized answer.

The difference changes how influence is measured. A buyer might ask an agent to identify software that meets five security requirements and integrates with an existing cloud platform. The agent can read documentation, news coverage, analyst material, and review content before returning three candidates.

The buyer may never see the ten sources behind that recommendation. A useful article can influence the result while receiving no referral traffic.

Google News still matters within this process because recent events often change enterprise risk. A product announcement can expand a vendor’s capabilities. A breach, lawsuit, acquisition, or service outage can alter the same vendor’s suitability.

However, freshness alone does not make a source decisive. Agents also need content that clearly identifies products, dates, limitations, and the relationship between a claim and its evidence.

Ambiguous pages create retrieval problems. A vendor that uses several names for one feature can look inconsistent. A page that hides essential specifications inside decorative interfaces may provide less usable evidence than a plain technical document.

The same problem affects news reporting. A headline may attract human attention but give an agent little context. Articles with precise attribution, dates, and links to primary material offer stronger evidence for comparison.

This does not mean companies should produce pages written only for machines. Buyers still read source material when a decision carries financial, operational, or professional risk. Clear writing serves both audiences.

It also does not mean traditional search has disappeared. Search engines remain useful for navigation, source checking, and finding recent coverage. Google News can become more valuable when a buyer wants to verify what an assistant summarized.

The real change is sequencing. An agent can form the first shortlist, while search becomes a validation tool later in the process. Vendors eliminated during the first step may never benefit from that later verification.

IDC’s findings therefore pressure marketing teams to treat product information as structured evidence. Claims need stable definitions, clear ownership, current dates, and supporting documentation. Contradictions between a sales page and a security document become easier for an agent to expose.

Independent material also gains weight. An agent comparing vendor claims can look for analyst research, credible reporting, technical evaluations, and customer evidence. Repetition across a company’s own pages does not create independent confirmation.

This is where knowledge quality becomes a purchasing issue. Teams must maintain consistent facts across product documentation, policy pages, help centers, and public announcements. Internally, a searchable AI knowledge base can help employees locate the evidence behind approved claims.

That internal discipline affects external trust. A salesperson who cannot explain where a claim came from weakens the buyer’s confidence. A representative who can retrieve the relevant benchmark, policy, or limitation becomes more useful.

For publishers, the incentive is similar. Original reporting, transparent attribution, and links to primary documents give an article lasting value beyond its initial audience. Summaries copied from other summaries become less defensible when agents compare sources.

The result is not the end of Google News. It is a change in the channel’s role. News becomes one input within a broader evidence system, and the visible click captures only part of its influence.

Vendor-Led Selling Is Losing Control of the First Evaluation

The central contest is vendor-led persuasion versus buyer-controlled analysis, not automated purchasing versus human judgment.

Sales teams once controlled access to important product details. Buyers often needed a representative to obtain configurations, technical answers, or a customized comparison.

Digital self-service reduced that control. AI agents reduce it again by helping buyers combine information from many sources without following the vendor’s preferred sequence.

A buying team can ask an assistant to extract contractual concerns, compare data residency options, or identify missing security disclosures. It can prepare questions before the first sales meeting and challenge an answer during the meeting.

That changes the representative’s role. Repeating information already available online adds little value. Clarifying uncertainty, resolving contradictions, and accepting accountability become more important.

Gartner’s survey provides evidence for this hybrid model. Its May 2026 buyer validation study covered 645 B2B buyers surveyed during August and September 2025.

Gartner found that 45% used generative AI during a recent purchase, mainly to gather information about products and vendors. Buyers used an average of seven information sources, showing that AI did not become their only channel.

The study also found that 67% preferred a sales-representative-free experience, while 70% preferred a completely digital, self-service experience. Those preferences appear to support the decline of vendor-led selling.

However, 69% preferred to validate AI-generated insights with a salesperson. That result provides the crucial counterweight to a simple automation narrative.

Buyers want control over when a representative enters the process. They do not necessarily want to remove that person from every consequential decision.

Gartner found that representatives remained especially important when buyers defined a business problem, identified a preferred supplier, gained internal support, and finalized a purchase. These are moments when context and accountability matter more than retrieval speed.

This creates pressure across several enterprise functions.

Marketing teams must publish material that survives comparison outside a carefully designed website journey. Product teams need consistent terminology and specifications. Sales teams must prepare to validate evidence rather than guard access to it.

Procurement also gains influence because AI makes systematic comparison easier. Teams can turn policy requirements into repeatable checks and apply them across a larger vendor set.

Forrester’s business buying research says procurement professionals act as decision-makers in 53% of business buying cycles. They now enter from the beginning rather than appearing only during contract negotiation.

Forrester also reports that a typical decision involves 13 internal stakeholders and nine external influencers. More complex purchases involve even larger groups.

AI can help those participants process information, but it cannot remove their competing priorities. Security may prefer limited data access. Finance may emphasize utilization. Operations may prioritize integration, while legal reviews contractual exposure.

An agent can organize these constraints and identify conflicts sooner. It cannot decide which stakeholder should accept a particular risk unless people explicitly delegate that authority.

This explains why the transition pressures poorly aligned vendors. A company may have accurate answers scattered across sales decks, support tickets, security portals, and employee knowledge. An AI buyer sees the gaps between those sources as uncertainty.

Internally connected vendors have an advantage. Employees can respond with consistent evidence because they can find the same approved information. A searchable knowledge base can help technical teams retrieve relevant records without reconstructing an answer from memory.

Still, internal organization does not guarantee external selection. Buyer-controlled evaluation makes direct comparison easier. When products appear similar, evidence quality, trial results, and implementation risk become stronger differentiators.

The first evaluation is moving beyond the vendor’s control. The opportunity for sellers remains, but it arrives later and demands more substance.

The Numbers Do Not Prove Autonomous Procurement Has Arrived

AI-assisted buying is widespread, but autonomous transaction authority remains a much higher threshold.

IDC’s 80% statistic combines assistance with task performance. Those categories can include very different levels of responsibility.

Summarizing a product page presents limited direct risk. Building a shortlist can materially shape a purchase. Submitting a request for proposal changes an external workflow, while approving payment can create a binding financial event.

Treating all four actions as equivalent would overstate current autonomy. The public IDC material establishes broad involvement, not the share of enterprise budgets controlled by independent agents.

Gartner’s findings reinforce that distinction. Buyers use AI for speed, yet 51% said they were more likely to encounter misleading information from generative AI. A nearly equal 49% associated misleading information with sales representatives.

Neither channel automatically earns trust. Buyers combine them because each can expose weaknesses in the other.

Forrester found that more than 60% of business buyers use some form of trial. These trials range from tailored sandboxes to usage-based evaluation periods.

A trial matters because it converts a claim into observable behavior. An AI assistant can summarize stated capabilities, but a controlled test shows whether the product works with the buyer’s data, controls, and workflows.

Trials also create evidence that agents cannot obtain from public sources. Performance in a specific environment, implementation effort, support quality, and user acceptance often become visible only during evaluation.

Security presents an even harder limit. An autonomous procurement agent needs access to vendor information, internal policies, identity systems, and possibly financial tools. Each connection expands the consequences of an error.

NIST’s 2026 agent security analysis found broad agreement that agents introduce novel threats. Respondents also agreed that familiar cybersecurity practices require adaptation for agent systems.

One risk is indirect prompt injection, where malicious instructions appear inside external content that an agent processes. A compromised page or document could attempt to redirect the agent, reveal sensitive information, or trigger an unintended action.

That threat has direct relevance to buying agents. Procurement research requires reading websites, emails, proposals, and attachments from parties outside the buyer’s organization. These are precisely the environments where untrusted instructions can enter a workflow.

An enterprise can reduce exposure by limiting permissions, separating research from execution, requiring approval for consequential actions, and recording what evidence supported each recommendation. These controls also make later review possible.

Data quality remains another constraint. A comparison generated from outdated specifications may appear precise while producing the wrong conclusion. An agent needs reliable dates, product identities, and rules for resolving inconsistent sources.

Bias can enter through source selection. Well-documented vendors may receive more attention than suitable vendors with weaker public material. Large companies may dominate retrieved evidence because they generate more coverage.

The agent’s ranking criteria can introduce further distortion. A request to minimize cost may overlook migration risk. A requirement for many integrations can reward breadth while ignoring the quality of each connection.

Human review does not automatically solve these problems. People can accept confident summaries without checking the underlying evidence, especially when the output confirms an existing preference.

This creates an automation paradox. The more efficiently an agent produces a polished recommendation, the easier it becomes to overlook uncertain inputs.

Companies should therefore distinguish between AI participation and AI authority. Participation includes research, extraction, comparison, and workflow preparation. Authority includes committing funds, accepting terms, or changing production systems.

The first category has clearly entered mainstream B2B buying. The second requires stronger identity, authorization, audit, and accountability mechanisms.

IDC’s finding remains important without claiming that procurement departments have become autonomous. The immediate shift is informational control. Agents increasingly influence what buyers see and which questions they ask.

The next phase depends on whether enterprises can convert that influence into reliable, governed action.

Three Signals Will Show Whether AI Buyers Gain Real Authority

Adoption counts matter less now than evidence of delegated authority, reliable outcomes, and vendor adaptation.

The first signal is the appearance of measurable transaction data. Enterprises and technology providers need to disclose which tasks agents complete, how often humans intervene, and what purchase values those systems handle.

A rise in autonomous transaction completion would strengthen the claim that AI agents have become buyers rather than research assistants. High override or correction rates would weaken it.

Transaction data should also separate routine renewals from new strategic purchases. Renewing a standardized subscription within a spending limit presents less uncertainty than selecting infrastructure for a major deployment.

The second signal is progress in agent identity and authorization. Organizations need ways to establish which agent is acting, whose authority it carries, which resources it can access, and how that authority can be revoked.

NIST has identified identity and authorization as central requirements for enterprise agent use. Clearer standards would make it easier for procurement platforms, vendor systems, and financial controls to recognize agents without granting excessive access.

Broad implementation of constrained permissions and auditable approvals would support more autonomous buying. Repeated security incidents or unclear liability would keep agents in an advisory role.

The third signal is how vendors redesign their digital channels. Watch for product data that becomes easier to verify, compare, and retrieve across documentation, security portals, marketplaces, and current reporting.

Vendors will also need to monitor whether AI systems describe their products accurately. A discrepancy between the official documentation and an agent’s answer can influence a shortlist before the company knows a buyer exists.

Google News will remain part of this evidence environment, especially when recent events affect vendor risk. Its commercial influence will become harder to measure if agents consume reporting without producing conventional referrals.

Successful vendors will not respond by flooding every channel with repetitive AI-generated content. Volume can increase contradictions, dilute attribution, and make meaningful evidence harder to identify.

They will publish precise claims, show limitations, maintain current documentation, and provide independent validation where appropriate. Sales representatives will then enter at high-trust moments with evidence that addresses the buyer’s actual uncertainty.

For enterprise buyers, the practical question is not whether to use an agent. Many already do. The question is which tasks deserve delegation and what evidence must remain visible to a human decision-maker.

Start by recording the sources behind important recommendations. Require approval before an agent sends commitments, accepts terms, or triggers payment. Test outputs against real requirements instead of judging their fluency.

Then watch the three signals: verified autonomous transactions, enforceable agent identity, and machine-readable vendor evidence. Together, they will show whether IDC’s 80% figure represents an early research habit or the foundation of a new purchasing system.

AI agents have already changed the route to an enterprise shortlist. The next decision belongs to buyers: will they keep these systems as fast research partners, or grant them authority over the purchase itself?