How Our Startup Grew Google Traffic 800% in 6 Weeks Using AI (And Replaced Our SEO Agency)

Six weeks ago, Duet was invisible on Google. We had four guides, zero blog posts, no structured data, and no third-party presence. Our Google Search Console dashboard was a flatline.

Today we're at 7,143 weekly impressions, 56 weekly clicks, a 0.78% CTR, and Google sent us a congratulatory email for hitting 80 clicks in 28 days. Our daily impressions went from 77 to 1,543 — a 20x increase.

We didn't hire an SEO agency. We didn't spend $5,000/month on a retainer. We used our own AI agent to do the entire thing — keyword research, technical SEO audits, content generation, structured data implementation, and weekly performance tracking.

Here's the exact playbook.

---

Where We Started: The Baseline

On February 10, 2026, we ran our first SEO audit. The results were humbling:

We had four Claude Code guides that were picking up some organic impressions, but that was it. No blog, no structured data, no third-party presence, no comparison pages. The site was functionally invisible for any non-branded query.

The "duet" brand query itself was a problem — Google's former Duet AI product, Duet Display, a dating app called Duet, and a PM tool called Duet were all competing for the same keyword. We were getting 904 impressions on "duet" with a 0.2% CTR because most of those searchers weren't looking for us.

The Decision: AI Agent Instead of Agency

A typical SEO agency charges $3,000–$7,000/month. For that, you get keyword research, a content calendar, some blog posts, and a monthly report. The actual value breaks down to about $800 in tool subscriptions, $2,200 in analyst hours, and $2,000 in content production.

We had something agencies don't: a persistent AI agent running on its own cloud server with access to DataForSEO, Google Search Console, Firecrawl, and the ability to write and ship code. Our agent could run keyword research, write content, generate structured data, create PRs in our codebase, and track results — all without context-switching between five different SaaS dashboards.

So we pointed our AI agent at the problem and let it work.

Week 1–2: Technical SEO Foundation

Before writing a single blog post, the agent identified our biggest technical gaps and shipped fixes as pull requests.

Structured Data (JSON-LD)

The agent added three layers of structured data:

1. Homepage: Organization + SoftwareApplication + WebSite schemas — telling Google exactly what Duet is, who built it, and how to categorize it
2. Guide pages: TechArticle + BreadcrumbList + conditional FAQPage schemas — marking up every guide with rich snippet eligibility
3. Guides listing: CollectionPage + ItemList schema — structuring the guides index for Google's crawler

This was PR #661, #663, and #715. Before these, Google had zero structured data to work with. After, every page was machine-readable.

AI Bot Crawling Rules

We updated robots.txt with explicit allow rules for AI crawlers — GPTBot, ChatGPT-User, PerplexityBot, ClaudeBot, OAI-SearchBot, and Google-Extended. Most sites either block these or don't address them. We wanted AI models to index our content for citation in AI-generated answers.

Title and Meta Optimization

The agent audited every guide title and meta description (PR #649, #651). Our Claude Code beginners guide had 1,928 impressions but a horrific 0.1% CTR — the title was 63 characters and getting truncated in search results. After shortening it to 49 characters and rewriting the meta description, CTR improved 233% over the following weeks.

Results after Week 2:

Five PRs shipped. The foundation was set.

Week 3–4: The Content Blitz

With the technical foundation in place, the agent shifted to content.

Keyword Strategy

The agent ran comprehensive keyword research using DataForSEO across our target segments (solopreneurs, startups, agencies) and personas (GTM, product dev, operations). It identified high-opportunity keywords where search volume existed but competition was low enough for a new domain to rank.

The critical insight: our Claude Code guides were driving 90% of all organic impressions. The "Claude Code for Founders" guide had the highest CTR of any page (3.63%) because it targeted a specific business use case rather than a generic technical tutorial. The business-angle content was outperforming generic technical content by 10x on CTR.

30 Blog Posts in One Day

On March 1, we published 30 blog posts. The AI agent researched topics, wrote content, optimized for target keywords, and submitted PRs. Topics ranged from "How to Automate Competitive Intelligence for Your Startup" to "Build and Deploy a Web App Using Only AI" — all targeting the long-tail keywords the research had identified.

The sitemap auto-generates from the content directory, so all 30 posts were immediately discoverable by Google's crawler. No manual sitemap submission needed.

Comparison Guides

We shipped five comparison guides (Duet vs Claude Code, Duet vs Codex, Duet vs Conductor, Duet vs OpenClaw, Duet vs Slack) targeting the "X vs Y" queries that AI models and search engines both prioritize. These got indexed within days — faster than any other content type we published.

Results after Week 4:

Google sent us a milestone email: "Congrats on reaching 80 clicks in 28 days."

Week 5–6: Optimization and Compounding

With the content published and indexing, the agent shifted to optimization.

CTR Optimization

The agent identified our biggest waste: the query "claude code terminal guide" was ranking at position 1 with 0% CTR. We were sitting at the top of Google getting zero clicks because the snippet didn't match searcher intent. The agent flagged this and queued a fix.

It also found that 22 of our 30 blog post titles exceeded 60 characters — meaning they'd be truncated in search results, killing CTR before anyone even read the headline.

AI Visibility Monitoring

Every week, the agent ran a structured AI visibility check — querying ChatGPT, Perplexity, and other AI models with category, comparison, and problem queries to see if Duet appeared in AI-generated answers.

Our AI citation rate went from 0% to 22% over the six weeks. We appeared in both brand queries. Category queries remained at 0% — the agent identified that every category query ("best AI workspace," "slack alternatives with AI agents") returned listicles from G2 and Capterra, platforms where we had zero presence.

After four consecutive weekly reports recommending it, we finally created G2 and Capterra profiles in week 5.

Backlink Strategy

David (our other co-founder) had previously built Aomni, which has significantly more domain authority than duet.so. The agent analyzed the backlink opportunity:

• Footer link on all Aomni pages — baseline link equity from a higher-DA domain
• Homepage banner ("From the makers of Aomni — meet Duet") — strongest single link
• Blog post on aomni.co ("Why we built Duet") — editorial link, passes the most authority

One strategic homepage link from a higher-authority domain is worth more than 100 footer links from an SEO standpoint.

What the AI Agent Actually Did

Let me be specific about the division of labor, because "we used AI" is vague.

The AI agent autonomously:

• Ran weekly Google Search Console performance pulls and generated reports
• Identified technical SEO gaps (missing structured data, truncated titles, missing meta descriptions)
• Generated JSON-LD schemas and submitted them as pull requests
• Conducted keyword research using DataForSEO APIs
• Wrote 30+ blog posts optimized for target keywords
• Tested AI citation rates across multiple AI models weekly
• Recommended strategic actions ranked by expected impact
• Tracked week-over-week metrics and flagged anomalies

Humans decided:

• Product positioning and messaging direction
• Which recommendations to act on (and when)
• When to create G2/Capterra profiles (we waited 4 weeks too long)
• Content strategy direction (doubling down on Claude Code content)
• Backlink strategy and domain authority plays

The AI generated the analysis and did the execution. Humans set the strategy and made the judgment calls. This is the right split — AI is excellent at systematic research, content production, and tracking. Humans are better at strategic prioritization and knowing when the data is pointing in the wrong direction.

What Worked

1. Technical SEO before content. JSON-LD, robots.txt rules for AI bots, and title/meta optimization gave us a foundation that amplified everything that came after. Structured data made our pages eligible for rich snippets. AI bot rules made our content available for citation.

2. The business-angle content strategy. "Claude Code for Founders" outperformed "Claude Code for Beginners" on CTR by 10x. Targeting a specific persona with a specific use case beats targeting a generic tutorial audience every time. Our best-performing content answered the question "how does this help my business?" not "how does this technology work?"

3. Comparison content gets indexed fastest. Our "Duet vs X" guides appeared in search results within days. Every other content type took weeks. If you need quick organic visibility, comparison content is the shortest path.

4. Weekly tracking with automated reports. The AI agent generated a structured report every week with consistent metrics, flagging regressions and opportunities. This cadence meant we caught the "position 1, 0% CTR" problem within a week of it appearing instead of finding it in a quarterly review.

5. Publishing velocity over publishing perfection. Thirty posts in one day is extreme. Most had titles that were too long. Some needed meta description work. But they started indexing immediately, and we could optimize after publication. Waiting for perfection would have delayed the entire growth curve by weeks.

What Didn't Work (Yet)

1. AI citation rate plateaued at 22%. Brand queries work. Category queries don't. Every non-branded AI search returns listicles from G2, Capterra, and established platforms. Without review site presence, there's a hard ceiling on AI visibility.

2. We waited too long on G2 and Capterra. The agent recommended creating profiles in week 1. We actually did it in week 5. Four weeks of missed opportunity for third-party discovery and AI citation improvement.

3. Brand name collision. "Duet" competes with Google's former Duet AI, Duet Display, a dating app, and a PM tool. Our homepage impressions dropped 45% in one week as Google resolved its own Duet AI brand confusion — and pulled our traffic down with it. There's no fix for this except building enough branded search volume that Google learns the intent.

4. Blog post titles over 60 characters. Twenty-two of our 30 blog posts had titles that get truncated in search results. This suppresses CTR from day one. We should have enforced a character limit in the content pipeline before publishing.

The Playbook (Steal This)

If you're a startup trying to grow organic traffic with AI, here's the sequence:

1. Audit first. Run a full technical SEO audit before writing any content. Missing structured data, broken meta tags, and missing AI bot rules are silent killers.

2. Add structured data on day one. JSON-LD for your homepage (Organization, SoftwareApplication), content pages (Article, TechArticle, FAQPage), and listing pages (CollectionPage, ItemList). This makes every page machine-readable.

3. Let AI bots crawl your content. Update robots.txt to explicitly allow GPTBot, PerplexityBot, ClaudeBot, and other AI crawlers. You want to appear in AI-generated answers.

4. Find your content wedge. Don't write generic content. Find the intersection of a keyword with search volume, a specific persona, and a specific use case. Our wedge was "Claude Code for business users" — not "Claude Code tutorial."

5. Ship comparison content early. "Your Product vs X" pages get indexed fastest and directly target high-intent queries. Publish 3–5 comparison guides in week one.

6. Publish in volume, optimize after. Get content indexed first. Fix titles, meta descriptions, and CTR issues once you have data. Perfection before publication is the enemy of growth.

7. Track weekly. Automated weekly reports with consistent metrics. Week-over-week trends show you what's working faster than monthly reports ever will.

8. Get on third-party platforms immediately. G2, Capterra, Product Hunt — these dominate AI citation results for category queries. Create profiles on day one, not week five.

The Numbers

Total cost: API fees for DataForSEO, Firecrawl, and AI model usage. Under $200 for six weeks of work that would have cost $18,000–$42,000 at agency rates.

The growth isn't stopping. Those 34 blog posts are still getting indexed. The structured data is compounding. And the AI agent is still running weekly audits, flagging opportunities, and shipping improvements.

We didn't replace our SEO agency. We never needed one.