The most significant recent change in B2B software buying is that the research layer moved a growing share of vendor discovery now happens inside AI interfaces before any search, review site visit or vendor contact.
Three changes get discussed constantly and are well understood. Buying committees grew. Cycles lengthened. Buyers do more research independently before contacting sales.
The fourth change gets far less attention and compresses the other three. The place where that independent research begins has shifted, and it produces a shortlist in under a minute.
What does the research process look like now?
A buyer describes their situation to an AI tool and receives a shortlist of named vendors with reasoning attached, then searches for the vendors they were given.
The sequence in practice:
1. The buyer describes their problem conversationally. “We’re a 60-person logistics company managing fleet maintenance in spreadsheets. It’s breaking. What software should we look at?”
2. They receive three to five named vendors, with justification. Not a page of links to evaluate a filtered set with reasoning.
3. They ask follow-up questions. Pricing, Suitability for their size. Comparison between two of the names. Integration with their existing stack.
4. Only then do they search for the specific vendors they were given.
5. They visit review sites to validate what they were told.
6. They contact two or three vendors.
The critical point is where filtering happens. It used to occur at step four, on a search results page, where being ranked eighth still meant being seen. It now occurs at step two, where being absent means not existing.
Why does this matter more than the other changes?
It matters more because it compresses the entire discovery phase and removes the partial credit that used to exist.
Compression: A buying committee that would have spent four weeks building a longlist now has one in ninety seconds. Everything downstream shortens with it.
No partial credit: In a ranked list of ten results, position eight produced some visibility, some traffic, some chance of being considered. In a generated answer naming four vendors, you are either one of them or you are absent. There is no position five.
Borrowed credibility: A synthesized recommendation reads as advice rather than advertising. The vendor named starts the evaluation with a trust advantage that paid placement cannot buy.
Invisibility of the loss: No system reports how often an AI named your competitors and not you. Traffic looks stable because the people who already know your brand still find you. The buyers who never learned you exist leave no trace in any analytics platform.
Is search dead?
No. Search volume remains substantial and most buyers still search but increasingly for vendors they were already given rather than to discover them.
This distinction matters for how you read your own data.
When a buyer receives a shortlist from an AI and then searches for those three vendors by name, the resulting traffic looks like healthy branded search. The vendors receiving it may conclude their brand marketing is working. The vendors absent from the shortlist see nothing no lost traffic to investigate, because it was never traffic.
What this changes about SEO?: Ranking for broad category terms delivers less than it used to, because fewer buyers are starting there. Ranking for branded and comparative terms delivers more, because that is where buyers arrive after the shortlist is formed.
What it does not change? The underlying discipline. Being genuinely useful, structurally clear and credibly cited is the same work it has always been. What shifted is which parts of that work pay off.
How much of B2B research actually happens this way?
The honest answer is that reliable, independent measurement does not yet exist, and anyone citing a precise figure is overstating what is known.
What can be said with more confidence?
- Adoption of AI assistants among knowledge workers has grown rapidly and continues to
- The behaviour is more common among technical buyers and younger decision-makers
- It is more prevalent in categories with many similar options, where filtering is genuinely hard
- It appears more often in early-stage research than in final vendor selection
The practical implication for a go-to-market leader: You do not need a precise market-wide figure. You need to know whether it is happening in your category, which you can determine in fifteen minutes by running the buyer prompts yourself.
If AI engines consistently name three of your competitors and not you, the percentage of buyers doing this is academic the exposure is real and it compounds.
What does this mean for pipeline?
Discovery problems surface in closed revenue three to four quarters after they begin, which is why they are usually diagnosed late
The mechanism
Month 1–3: Fewer new buyers learn you exist. Nothing visible changes your pipeline is still being fed by discovery that happened earlier.
Month 4–6: Top-of-funnel volume softens. Usually attributed to seasonality, market conditions or a specific campaign underperforming.
Month 7–12: Closed revenue reflects it. By this point the cause is a year old and the analysis is archaeological.
The compounding element: Brands that get named accumulate mentions, reviews and third-party coverage, which makes them more likely to be named next time. Brands that are absent accumulate nothing. The gap widens without either party doing anything differently.
This is why the honest framing for a leadership team is not “we should optimize for AI search.” It is there is a channel where we may be losing to competitors, it does not appear in any of our reporting, and the check takes fifteen minutes.
What should a go-to-market leader do about it?
Run the diagnostic first. Do not commission a strategy for a problem you have not confirmed you have:
Step 1 Check (this week, 15 minutes)
Open ChatGPT and Perplexity in fresh sessions. Ask for the best tool for your ideal customer’s specific situation. Ask for alternatives to your largest competitor. Ask which options suit a company of your customer’s size. Record which brands appear.
Step 2 Interpret:
If you appear consistently, this is not your most urgent problem. Spend your attention elsewhere and recheck quarterly.
If competitors appear and you do not, you have confirmed exposure in a channel your reporting does not cover.
Step 3 Diagnose the cause:
In most cases it is one of three things: no comparative content exists about you anywhere, nothing outside your own domain corroborates your claims, or your basic facts are not stated in a form that can be extracted.
Step 4 Fix content type, not content volume:
The most common error is responding by publishing more. If twelve months of content did not produce this visibility, more of the same will not either. The gap is comparison content, extractable facts and third-party presence.
Step 5 Measure monthly:
Fifteen to twenty fixed prompts, run in fresh sessions, logged. Without a baseline you cannot tell whether anything is working, and that uncertainty is what causes teams to abandon the effort in month three.
What this is not?
Three overstatements worth resisting, including from vendors selling the solution:
It is not the end of SEO: The foundations are shared and largely unchanged. A site with poor technical health and thin content fails at both.
It is not fully understood: These systems are opaque, they change without notice, and much of what circulates as methodology is inference from limited observation. The measurable parts are genuinely measurable; the mechanics are not fully known.
It is not equally urgent for everyone: In categories with transactional purchases, strong existing brand recognition or heavy relationship-led selling, the exposure is lower. Run the check rather than assuming either way.
The reasonable position is that this is a real and growing channel where most companies have no visibility into their own position and that the cost of checking is fifteen minutes.
Frequently Asked Questions:
What percentage of B2B buyers use AI tools for research?
No , reliable independent measurement exists, and precise figures circulating should be treated sceptically. What matters practically is whether it is happening in your category, which you can test directly.
Less so at final selection, where procurement processes and analyst relationships dominate. More so in early research, where individual stakeholders form initial views before any formal process begins.
Should we reduce SEO investment
No, Reprioritize within it. Broad category terms deliver less; comparative content, extractable facts and third-party presence deliver more.
How do we know if we’re affected?
Run five buyer-intent prompts through ChatGPT and Perplexity in fresh sessions. If competitors are consistently named and you are not, you are affected.
Is this a permanent change or a phase?
The interface may change; the underlying behaviour buyers wanting a filtered, justified shortlist without doing the filtering themselves has been consistent for decades. Review sites and analyst reports served the same need.
How long does it take to fix?
Specific, high-intent queries typically move within eight to fourteen weeks. Broad category queries take six to twelve months, and in entrenched categories may not be winnable.