3 Years Since ChatGPT: What AI Actually Changed in SMB Marketing

3 Years Since ChatGPT: What AI Actually Changed in SMB Marketing

ChatGPT launched on November 30, 2022. That was a Wednesday. I remember where I was when I first used it — a Monday morning in early December, sitting in the Longview office, running a test prompt to see if it was real.

It was real. It was also nowhere near as transformative as the first week of coverage suggested, and nowhere near as limited as the backlash suggested six months later.

Three years in, I want to write an honest retrospective. Not a celebration of AI, not a warning about AI. An accounting of what changed, what did not, and what is just now beginning to move.

---

The Framework: Hyped vs. Real vs. Still Ahead

I’ll organize this by category because the hype, the reality, and the horizon are different in each area. The comparison table is at the end. The reasoning is throughout.

---

What Won: Three Things That Actually Changed

AI Content at Volume

This one is real and durable. Before November 2022, producing a 1,500-word first draft required a writer, a brief, a timeline, and a budget. Today, a well-structured prompt produces a first draft in under two minutes.

What this changed in practice: a marketing team or agency that previously published four blog posts per month can now publish 12 to 16 without adding headcount, using AI for first drafts and structural outlines while humans handle voice, fact-checking, and editorial judgment. Email sequence drafts that once took half a day take 20 minutes. Social post calendars that once required a dedicated coordinator can be batch-drafted in an hour.

The businesses that got this right treated AI as a production tool with a human review layer. The businesses that got it wrong removed the human layer and published generic, brand-dead content that hurt their search presence and credibility.

The net result after three years: content volume is up across the industry. Content quality standards are polarizing — there is more excellent content from teams using AI well, and there is significantly more garbage from teams using it without oversight. Google’s helpful content system is responding accordingly.

AI-Assisted SEO

Keyword research, content gap analysis, meta description generation, title tag optimization, FAQ generation, and competitor content summarization are all faster and cheaper with AI assistance.

The specific workflow change that matters most: a full keyword universe that once took 4 to 6 hours to build (query expansion, clustering, intent classification, opportunity scoring) now takes 45 to 90 minutes with AI handling the expansion and clustering while a human handles the strategic prioritization.

Content audits that required reading and scoring 200 existing pages take 60 percent less time when AI handles the initial quality scoring and the human reviews the flagged items.

This is not a replacement for SEO expertise. It is a force multiplier for it. A skilled SEO practitioner with AI tools produces more work at higher quality than without them. An unskilled operator with AI tools produces bad work faster.

AI Customer Service

Chat bots, voice bots, and AI-handled customer service workflows are running at real businesses at real scale. The use cases that actually work: answering FAQs about hours, pricing, and availability; routing inquiries to the right department; handling initial qualification questions before connecting to a human; and following up on service requests after hours.

What changed for local businesses specifically: 24/7 availability for basic inquiries. A small business that could not staff a customer service line at 11pm on a Sunday can now have an AI bot handle “what are your hours,” “do you accept my insurance,” and “can I schedule an appointment for Thursday” — and capture that lead before the prospect moves to a competitor.

The failure mode is over-deployment. Businesses that replaced human customer service entirely with AI bots in the first wave of adoption found that complex inquiries frustrated customers. The model that works is AI for basic routing and information, humans for anything requiring judgment or relationship.

---

What Disappointed: Three Things That Underdelivered

Fully Automated Marketing Campaigns

The promise: you describe your business, your goals, and your audience. AI builds, launches, and optimizes your full marketing campaign with no further human input.

The reality: it does not work at the level of quality that produces sustainable marketing results. Not because the AI cannot draft the components, but because marketing strategy requires judgment that AI does not exercise reliably.

What problem is this campaign actually solving for this customer at this moment in their journey? When should we pull back spend because the offer is wrong, not because the bid is too low? What does a particular client relationship need right now that a campaign cannot provide?

These are judgment questions, not data questions. AI can present data that informs the judgment. It does not replace it.

After three years, the best AI-assisted campaigns are still directed by human strategists who are using AI to go faster and cover more ground, not to go unsupervised.

AI Ad Creative Replacing Human Designers

AI image generation — Midjourney, DALL-E 3, Adobe Firefly, Stable Diffusion — is genuinely useful for ideation, mood boards, and concept exploration. It is not where client-facing ad creative for East Texas businesses lands.

The specific failure mode: AI-generated imagery tends toward the aesthetically averaged. It produces images that look like everything and nothing. For a brand like a Longview dental practice or a Tyler law firm trying to communicate a specific, trustworthy, local identity, the AI average is the enemy.

The designers who use AI for rapid concepting and research, then produce final work with human craft and brand intelligence, are doing better work faster. The designers who use AI as a replacement for the full creative process are producing content that looks like it came from a template. Clients notice the difference even if they cannot articulate why.

AI Replacing Account Strategy

This one carried the highest hype and has the largest gap from reality. The prediction was that AI would analyze a business’s marketing data, identify opportunities, build a plan, and execute it with minimal human input.

The reality: AI is very good at analysis and very bad at strategic prioritization for a specific client in a specific context. It can generate a comprehensive list of things you could do. It cannot tell you, with confidence, which three of those things matter most given this client’s cash position, their team’s capacity, their seasonal business cycle, and the competitive pressure they are facing in their specific market this quarter.

That judgment is not in the data. It is in the relationship. It is in knowing the client well enough to understand the difference between what they are asking for and what they actually need. Three years in, that is still a human function.

---

What Is Just Now Emerging: Three Things That Will Be the Real Story

Agent-Based Commerce

The conversation around AI shifted significantly in 2025 from “AI as a tool you use” to “AI as an agent that acts on your behalf.” The distinction matters enormously for SMB marketing.

In a world where consumers use AI agents to research, compare, and transact — an agent that a consumer instructs to “find me the best orthodontist in Longview that takes my insurance and has availability in the next two weeks” — your business needs to be visible to that agent, not just to a human doing a Google search.

Agent visibility requires structured data that an AI can parse and act on. It requires consistent business information across all platforms that an AI reconciles when building its recommendation. It may eventually require API-accessible booking systems that an agent can interact with directly without a human in the loop.

This is not fully live in consumer behavior yet. But the infrastructure investments being made now — in schema markup, in business information consistency, in booking system API availability — will determine which local businesses benefit when agent-based commerce becomes mainstream in the next 18 to 36 months.

GEO as a Discipline

Generative Engine Optimization did not exist as a named practice in November 2022. As of April 2026, it is a specific service with a specific methodology, a specific set of tools, and measurable outcomes.

The businesses that invested in GEO foundations in 2024 and 2025 — structured content with explicit author attribution, FAQ schema, Organization and LocalBusiness schema, citation footprints across authoritative platforms — are appearing in Google AI Overviews, Perplexity answers, and Gemini responses in 2026. Those that have not invested are invisible to AI-generated answers on queries where their competitors are getting cited.

GEO is early enough that the advantage is still available for businesses willing to move. It is late enough that the lead is compounding for those who started a year ago. The window for easy entry is narrowing.

AI as Orchestration Layer

The most significant emerging pattern is not AI as a tool for any specific task. It is AI as the coordination layer between systems.

An AI system that monitors your Google Ads performance, identifies budget pacing anomalies, pulls the relevant data from your CRM, sends an alert to the account manager with a drafted response recommendation, and logs the event in your project management system — without any human initiating the workflow — is not science fiction. It is running in early production environments now.

For Starfish, this is the direction the AI Employee system is building toward. Not a chatbot. Not a content generator. An operational layer that observes, coordinates, and surfaces decisions to humans who then act. The human judgment is preserved. The mechanical coordination is automated.

This is the version of AI that actually changes how agencies and SMBs operate. It is just now becoming practical at the business size we serve.

---

The Honest Assessment: Three-Year Comparison Table

Category	What Was Hyped	What Actually Happened	Where It Stands in 2026
Content production	AI replaces writers entirely	AI produces first drafts; humans edit, fact-check, and add voice	Mature, stable — standard workflow tool
SEO	AI handles all keyword research and optimization automatically	AI speeds up research and analysis; strategy still requires human judgment	Mature, stable — integrated in most agency workflows
Customer service	AI fully replaces human CS teams	AI handles Tier 1 routing and FAQ; humans handle complex issues	Mature, stable — most effective in hybrid deployment
Ad creative	AI-generated images replace design agencies	AI useful for ideation; human craft required for final brand-specific work	Partially delivered — ideation useful, execution not replaced
Campaign automation	AI runs full campaigns with no human oversight	Requires significant human direction; autonomous campaigns underperform	Underdelivered — human direction still required
Account strategy	AI replaces marketing strategists	AI produces analysis; humans provide judgment and client relationship context	Significantly underdelivered
GEO	Not on radar in 2022	Emerged as distinct discipline; early movers seeing citation results	Just now maturing — 12-18 month advantage window
Agent commerce	Not on radar in 2022	Infrastructure phase — booking APIs, schema, consistency	Just now emerging — 18-36 month horizon
AI orchestration	Vaguely promised as “autonomous agents”	Early production environments at larger organizations	Just beginning — most significant long-term change

---

What This Means for SMBs in East Texas and Shreveport-Bossier

The businesses in Longview, Tyler, Marshall, and the Shreveport-Bossier area that have benefited from the last three years of AI development are the ones that used it for what it is good at: faster first drafts, faster research, 24/7 basic customer service, and improved consistency across marketing operations.

The businesses that are positioned for the next three years are the ones building the infrastructure that AI agents will need to surface and transact with them: complete structured data, consistent business information across platforms, validated schema markup, and author attribution that builds AI citation credibility.

The story of the next three years will not be about any single AI tool. It will be about which businesses built the right foundation to participate in how AI systems discover, evaluate, and recommend local businesses.

That foundation is available to build now. It does not require a large budget. It requires consistent, deliberate action over 12 to 18 months.

Three years from today, the retrospective for April 2026 to April 2029 will be about who built that foundation and who did not. I want our clients to be in the first group.

---