The most useful finding in the Philadelphia Fed’s new small-business evidence is not that AI use is growing. It is that reported productivity gains have not yet translated into broad changes in sales, labor costs, spending on outside services, or quality for most AI-using firms in Delaware, New Jersey, and Pennsylvania. For people doing GEO work, that is a necessary correction to the lazy equation of AI adoption with business performance.
The Federal Reserve Bank of Philadelphia found that 71% of Third District firms using AI reported increased productivity. Yet more than 70% reported no change in labor costs, outside-service spending, quality of goods or services, or sales attributable to AI. Crucially, the survey did not ask firms how they measured their productivity gains. philadelphiafed.org The evidence supports a narrower claim than many AI narratives make: firms perceive operational help. It does not establish that AI has broadly improved commercial outcomes.
Adoption is lower, and integration is shallower, than the headline suggests
The Philadelphia Fed analysis draws on the 2025 Small Business Credit Survey’s optional AI module. The survey defines a small business as a firm with one to 499 employees; responses were collected from September 3 to November 14, 2025. The optional module received 5,248 responses nationally and 679 from Third District states, with data weighted to represent small businesses in the United States and the Third District. philadelphiafed.org
At the time of the survey, 37% of Third District firms reported using AI, compared with 46% nationally. Among the region’s adopters, 57% characterized their use as experimentation and just 3% said AI was fully integrated into business operations. Nearly half of Third District firms reported no AI use and no plans to adopt it, compared with one-third nationally. philadelphiafed.org
That matters because “AI adoption” combines very different states: a business trialling a tool, a team using it in selected tasks, and a firm that has integrated it into operations. The Fed’s figures show that the first category dominates in this regional sample. An adoption percentage alone therefore says little about whether AI has changed a firm’s workflow, customer experience, or results.
The regional gap may reflect industry composition, but the Fed presents that as a potential explanation rather than a settled cause. Its Third District sample included more manufacturing than the national sample, 12% versus 4%, and lower shares of nonmanufacturing goods production and services. The Fed also notes that production workers had among the lowest generative-AI exposure scores of major occupation groups in its prior report, which suggests fewer generative-AI use cases for manufacturing firms than for service firms. philadelphiafed.org
Writing and marketing lead, but that is not a revenue result
For GEO practitioners, the task-level data deserves more attention than the broad adoption number. Among Third District AI adopters, 79% used AI for writing or marketing. Individual productivity work such as notetaking and summarizing was cited by 56%, while 51% used AI for planning or analysis such as research. Roughly one-third reported administrative functions or customer service, 16% process automation, and 12% coding-related tasks. philadelphiafed.org
Writing and marketing being the leading category is compatible with AI being useful in a GEO workflow. But it is not evidence that a business has gained visibility, qualified demand, sales, or a durable position in AI-mediated discovery. The same Fed results say that most firms reported no AI-related change in sales. philadelphiafed.org
That distinction should affect how small-business teams describe the work. Producing more drafts, summaries, research notes, or marketing variations is an operational output. It can be valuable, but it is not the same thing as a business outcome. The evidence does not tell us whether the firms reporting productivity gains achieved them through faster completion, more output, less effort, or some other measure. The survey did not ask. philadelphiafed.org
So the appropriate analytical stance is not to dismiss the productivity reports. It is to avoid extending them beyond what was measured. For a small business investing in GEO, the relevant question is not simply whether an AI tool makes content work faster. It is whether the specific workflow improvement changes the commercial metric the business intended to influence. The Philadelphia survey cannot answer that question for an individual GEO program, and it does not claim to.
What the evidence says about labor savings
The results also challenge a common assumption that productivity gains quickly show up as lower labor spending. The Philadelphia Fed reports that few Third District firms saw decreases in labor costs or spending on outside services because of AI, despite the reported productivity increases. It interprets this as evidence that AI had not significantly reduced regional small businesses’ spending on internal or external labor at the time of the survey. philadelphiafed.org
National Small Business Credit Survey responses on employment point in the same direction: 79% of firms reported no AI-related employment change, while 11% reported decreases and 10% reported increases. The Fed says the broader empirical evidence on AI adoption and labor demand remains mixed, with no clear causal link between AI and job losses in AI-exposed occupations or junior hiring. philadelphiafed.org
For a GEO team or external provider, this weakens any cost-saving pitch that assumes AI replaces work by default. The available regional evidence suggests that productivity may be absorbed into existing work rather than immediately reducing headcount, agency spend, or other external-service costs. That is not a failure of the technology; it is a reminder that gains can be real without appearing in the financial line item a buyer expects.
The measurement gap is the actual operating problem
The Fed’s finding should be read alongside the New Jersey State Policy Lab’s literature overview, which emphasizes both potential benefits and significant barriers. It identifies operational efficiency and customer engagement as frequently observed benefits in the literature, while noting that much of the academic literature was conducted outside the United States. policylab.rutgers.edu
The same overview identifies financial resources and technological skill as frequently observed adoption barriers, alongside poor IT infrastructure, perceived investment risk, lack of guidance, and concerns about data privacy and security. It also identifies difficulty accessing quality data as a substantial barrier to AI integration. policylab.rutgers.edu Those constraints help explain why a tool that assists with writing or research may be easier to trial than a system that changes core operations.
The Philadelphia Fed’s national sample also found that more than 20% of AI-using firms cited challenges including accuracy, adapting AI to business needs, data security and privacy, training time, and costs. philadelphiafed.org None of those are abstract implementation concerns for GEO. They bear directly on whether a small business can turn AI-assisted research, writing, and analysis into a repeatable process with standards it can trust.
What this changes for GEO work
The Fed data makes a modest but important case for discipline. AI is already most common among these adopters in work adjacent to marketing, writing, research, and individual productivity. philadelphiafed.org That makes GEO a plausible area for experimentation. It does not make broad claims of sales growth, expense reduction, or transformation credible on its own.
- Separate adoption from integration. A team using AI for selected drafts is not necessarily operating an integrated AI system.
- Separate productivity from performance. Self-reported productivity is the strongest reported effect in the regional survey; most core performance measures did not change.
- Define the outcome before claiming value. The survey’s lack of a productivity measurement question is a warning against relying on an undefined gain.
- Keep uncertainty visible. The evidence is regional, survey-based, and reflects the fall of 2025; it is not a causal estimate of what AI will do for every small business.
Small-business AI evidence is still developing. The Rutgers overview notes gaps created by inconsistent definitions of small businesses and a lack of refined research on women- and minority-owned firms rather than research that conflates those businesses with entrepreneurship and startups. policylab.rutgers.edu The Philadelphia Fed evidence is valuable precisely because it does not erase that uncertainty. It shows operational optimism, limited integration, and little broad reported change in core outcomes so far. That is a far more useful starting point for GEO planning than an adoption headline.