The New York Fed’s latest small-business evidence cuts against the simplistic claim that AI adoption automatically means fewer jobs. In the 2025 Small Business Credit Survey, small firms using AI were more likely than non-users to expect employment and revenue growth over the following year, while 77% reported no change in current labor costs from AI use. libertystreeteconomics.newyorkfed.org
For people doing GEO work, that distinction matters. The survey does not establish that AI has already delivered broad, measurable commercial gains, nor does it show that AI is causing firms to add staff. It shows an association between adoption and more positive expectations. That is useful dated evidence for planning, but it is a poor basis for treating “AI adoption” as a substitute for measuring whether a new search workflow, content process, or technical investment is producing results.
The evidence is about expectations, with a limited operating baseline
The analysis, published October 8, 2026, uses the 2025 Small Business Credit Survey to examine twelve-month-forward employment and revenue expectations among U.S. small businesses. The survey’s AI module covered 5,248 employer firms and was fielded between September and November 2025; responses were weighted to match the national population of employer firms. byteseu.com
Its adoption figures are substantial: 46% of firms with at least one employee said their business or employees were using AI tools, and another 15% said they planned adoption in the next twelve months. Among AI users, 63% said AI was somewhat or very important to production, while 51% said it was partially or fully integrated into business processes. libertystreeteconomics.newyorkfed.org
Those figures should not be read as a uniform maturity signal. “Using AI,” saying it is important to production, and partially or fully integrating it are separate survey measures. A GEO practitioner should make the same distinction inside an organisation. A team that occasionally uses AI for research or drafting is not necessarily operating a repeatable system for source selection, content quality control, technical implementation, or outcome measurement.
The near-term operating results in the survey are notably more restrained than the forward outlook. Seventy-seven percent of respondents reported no change in current labor costs from AI use, and only 31% reported increased sales from their firm’s AI use. byteseu.com That is the central corrective to AI job-loss narratives and to equally unhelpful ROI narratives: adoption can be real, integration can be meaningful, and present-day business effects can still be limited or uneven.
AI users expect more hiring, not less
Among employer firms currently using AI, the net expectation of higher employment over the next twelve months was 33 percentage points. For non-users, it was 15 percentage points. In this survey, “net expectation” means the share expecting an increase minus the share expecting a decrease. libertystreeteconomics.newyorkfed.org
That raw gap could reflect differences between adopters and non-adopters rather than AI use itself. The authors report that younger, more profitable firms and firms with more than ten employees were likelier to have more optimistic employment expectations. After accounting for firm, owner, and location characteristics, AI users still had a 13-percentage-point higher net probability of expected employment growth. After also accounting for past employment changes, the difference remained positive at 10 percentage points. byteseu.com
This is more useful than a headline about AI “creating jobs,” but it is not proof of causation. The authors themselves frame their conclusion as an association, and say whether AI-adopting firms actually achieve the stronger employment and revenue outcomes they anticipate is a subject for future research. byteseu.com The available evidence therefore supports a narrow statement: surveyed adopters were more optimistic about hiring after the analysis accounted for several observable differences. It does not settle what AI will do to employment.
For GEO leads, the practical implication is to reject an either-or staffing model. The relevant choice is rarely “AI or people.” It is whether AI changes the mix of human work: research, editorial judgment, subject-matter contribution, publishing operations, measurement, and technical remediation. The survey cannot tell us which work changes in a GEO team, but it does offer no support for assuming that adoption necessarily removes the need for people.
Revenue optimism is stronger than reported sales gains
The revenue pattern is similar. Among current AI users, the share expecting revenue to rise exceeded the share expecting it to fall by 48 percentage points. Among non-users, the equivalent net expectation was 21 percentage points. libertystreeteconomics.newyorkfed.org
After accounting for owner and firm characteristics plus revenue changes over the prior twelve months, AI users retained a 14-percentage-point advantage in net expected revenue growth over non-users. byteseu.com Again, that is a better-controlled association than the initial comparison, but it remains an expectation measure rather than an observed future result.
The most interesting segment is not simply firms in good shape. Among owners reporting fair or poor current financial conditions, the net expected-revenue-growth difference between AI users and non-users was 24 percentage points; it was 15 percentage points among owners reporting good conditions. The analysis says these differences were statistically significant. byteseu.com
For a small business trying to improve visibility in AI-mediated search, this suggests a more disciplined framing of the investment. AI tools may be appealing not only to firms already expanding, but also to firms trying to overcome constraints. Yet optimism under constraint is still not measured revenue. A GEO programme should therefore separate leading operational measures from commercial outcomes rather than collapse them into an “AI impact” claim.
Technology friction may be where the optimism concentrates
The survey asked about operational challenges in utilizing technology, including websites, social media, ecommerce, and cybersecurity. Among AI users, firms currently reporting such technology challenges had a 16-percentage-point higher net expectation of revenue growth than AI users that did not report those challenges. byteseu.com
The authors interpret this as suggesting that AI may improve revenue expectations by helping firms overcome internal technological constraints that would otherwise limit growth. byteseu.com That is a plausible working hypothesis, not a verified causal mechanism. The source does not identify which tools produced the effect, how firms used them, or whether the anticipated gains occurred.
Still, it is the most relevant finding for GEO operations. Search visibility work often fails before the model, prompt, or content format becomes the issue: the business may have weak website operations, inconsistent publishing, limited technical capacity, or unclear ownership of digital channels. AI may help a team work around some of that friction. It cannot make unmeasured activity into business performance.
What to take from the survey
The New York Fed evidence is a useful antidote to broad employment claims because it distinguishes current reported changes from expected future changes. It also shows why the word “adoption” needs qualification. The same survey contains widespread reported use, limited reported sales improvement, mostly unchanged labor costs, and notably stronger employment and revenue expectations among adopters. libertystreeteconomics.newyorkfed.org
- Do not use AI adoption as evidence that jobs have already disappeared; the surveyed firms more often expected employment growth, and most reported unchanged current labor costs.
- Do not use positive expectations as evidence of delivered commercial ROI; only 31% of respondents reported increased sales from AI use.
- Assess GEO work as an operating system: identify the constraint, define the human role, and distinguish process progress from revenue or employment outcomes.
The evidence is fresh and specific, but its limit is equally important. It measures reported use and expectations in a defined small-business survey. It does not establish a universal employment effect, prove that AI caused the expectations, or demonstrate that the expected growth arrived. That restraint is precisely what makes it more valuable than a generic story about AI replacing work.