The Shift
What changed from August 2026.
Between the August 2026 and September 2026 editions, 4 figures or themes were added, 2 changed and 1 were removed. All of it is listed here, because a brief that revises itself silently is not a source anyone should build an argument on.
- G2’s buyer-first findings on AI-first research, changed vendor choice, unfamiliar-brand discovery, decision speed and verification behavior
- The six-stage new buying journey: Define the need, Discover, Compare, Verify, Shortlist and Act
- The five company tests and eight-outcome Shift scorecard
- Explicit evidence boundaries separating reported findings from universal claims
- The report is now structured around buyer behavior and commercial decision points rather than adoption figures alone
- Muck Rack research is labeled vendor-published and citation-overlap studies are no longer compressed into one synthetic range
- August headline figures that are now preserved in the August 2026 archive
Buyers no longer start with ten blue links.
They ask an engine, and the engine answers with a shortlist. Either a company is in that answer or a competitor is. The 13 figures below are why, and each one states its publisher and its date so it can be checked rather than taken on faith.
of surveyed B2B software buyers start research with an AI chatbot more often than with Google, up from 36% seven months earlier.
How this can be misread: This applies to G2’s surveyed B2B software-buying population, not all industries or all buyers.
of surveyed software buyers said information surfaced by an AI chatbot led them to choose a different vendor than they initially expected.
How this can be misread: The finding is self-reported influence, not a controlled causal effect.
of surveyed software buyers said AI chatbots made their purchasing decision faster: 37% said much faster and 46% somewhat faster.
of surveyed B2B software buyers encounter AI inaccuracies weekly or more often.
How this can be misread: This is specific to the surveyed B2B software-buying population and does not measure every AI answer environment.
AI is changing the shortlist
The first commercial effect is not a click. It is whether a company enters the buyer’s consideration set at all.
of surveyed software buyers purchased from a vendor they had not previously heard of after AI surfaced information about it.
How this can be misread: The result is self-reported and applies to G2’s surveyed software buyers.
of surveyed software buyers said they think more highly of a vendor when an AI chatbot mentions it in a recommendation.
How this can be misread: This reports perceived influence, not a measured change in purchase probability.
of surveyed B2B software buyers said a citation from a review site is the single most confidence-inspiring signal in an AI answer.
The answer may end the visit
When an AI summary satisfies the research step, being cited becomes a consideration and reputation outcome as much as a referral-traffic outcome.
the share of visits with a traditional search-result click when a Google AI summary appeared versus when one did not.
Basis: 68,879 Google searches by 900 consenting U.S. adults in March 2025
How this can be misread: The study should not be generalized to every country, engine or later interface version.
of visits containing a Google AI summary produced a click on a source cited inside the summary.
Basis: 68,879 Google searches by 900 consenting U.S. adults in March 2025
of tracked commercial queries returned an AI Overview in BrightEdge’s measurement, a 58% relative increase in one year.
How this can be misread: This is not 48% of every Google query worldwide; prevalence varies by dataset, query class, country and window.
The buying journey now has an Act gate
AI-mediated buying moves through define, discover, compare, verify, shortlist and act. The final step is whether a person or agent can complete the commercial next step.
year-over-year growth in U.S. retail visits referred by generative AI in the first quarter of 2026.
Basis: More than one trillion U.S. retail visits
How this can be misread: The growth begins from a small base and is not equivalent to growth in Google organic traffic.
higher conversion for AI-referred retail visits than non-AI traffic in March 2026, reversing from 38% worse one year earlier.
Basis: More than one trillion U.S. retail visits
How this can be misread: The comparison group includes multiple non-AI channels.
of analyzed AI citations came from earned media broadly defined, including journalism, academic, government, NGO, encyclopedic, social and third-party corporate sources.
Basis: More than 25 million cited links across ChatGPT, Claude and Gemini in 17 industries
How this can be misread: Muck Rack sells AI-visibility software, and its broad earned-media definition is not synonymous with journalism.
Every source, and how independent it is.
Research in this category is published largely by companies selling into it. That does not make it wrong, and on several questions it is the only data that exists — but a reader deserves to know which findings come from a disinterested party and which come from a firm with a product to sell. Each figure above carries its tier. Here is what the three tiers mean.
Independent research
1 sourceSurvey organizations, newsrooms, infrastructure operators and academic bodies with no product to sell in this category.
Pew Research Center
Platform disclosure
1 sourceFigures published by the companies operating the engines or measuring the traffic. Authoritative on their own systems, and self-interested about the size of them.
Adobe Digital Insights
Vendor research
3 sourcesStudies published by firms that sell software or services in the search and answer-optimization market. Frequently the only data that exists on a question, and worth reading with the seller's interest in view.
BrightEdge · G2 · Muck Rack Generative Pulse
Figures are third-party published research, dated as reported by each source and cited so they can be checked. They are not AIGNCI measurements. Figures from different studies are not combined as if they measured the same thing. Every finding states its scope, source class, commercial interest and material boundary where disclosed.
Other editions
August 2026
Archived as published · 17 figuresSeventeen figures on how buyers now find companies, each carrying its publisher and date. This edition widens and re-sources the citation-overlap finding printed in July, and adds three findings on where AI citations actually come from.
July 2026
Archived as published · 14 figuresThe first edition, establishing that buyers had moved to AI-mediated research. Its citation-overlap finding was corrected in the August edition and is preserved here as published.
Tell us what you are building.
If the question is about discovery, reputation, a new site, or a market move, send the context. AIGNCI will tell you whether an Audit, a build, or a more focused engagement is the right starting point.