5 concrete lessons from an AI agent that generated 35% of leads via AI Search
Summarize this article with AI
Is SEO dead? No, it’s become an engineering problem
I’m going to tell you something agencies hate hearing. AI Search didn’t kill SEO. It turned it into an engineering problem.
On July 7, Samanyou Garg, CEO of Writesonic, dropped numbers that shook the Search Engine Journal webinar. 2.5% of Writesonic’s leads came from AI Search a year ago. 35% in March. Yes, you read that right. More than a third of prospects arrived through an AI-generated response, not a classic blue link.
This isn’t a passing spike. It’s a fundamental trend I’m tracking across my e-commerce clients. Product pages, buying guides, comparisons get cited in responses from Bard, ChatGPT, SGE. And without an engineering system behind it, you control nothing.
Garg laid out 5 field lessons, with 150,000 citations analyzed and a 6-step loop. I’ll translate them into concrete actions for your e-commerce site.
Writesonic’s team applies this 6-step loop to every page to sustain AI search visibility. Each step feeds into the next, creating a continuous improvement cycle.
The 6-step engineering loop for AI search success
From monitoring to learning, an iterative process to maintain citations
Lesson 1: Build a 6-step loop, not a one-time audit
AI Search rewards those who iterate fast. Not those who optimize a page and move on. The Writesonic team set up a 6-step loop applied to every published page.
1. Monitor: agents continuously scan AI responses for 1,200 target keywords. 2. Measure: they identify which pages are cited, which disappear, which competitors take their place. 3. Prioritize: a score combines search volume and proximity to purchase intent. 4. Act: refresh content, modify semantic markup, contact a third-party site. 5. Verify: 48 hours later, an agent confirms if the citation returned. 6. Learn: each iteration feeds a dashboard that refines optimization rules.
« AI search didn’t necessarily kill SEO, but it turned it into an engineering problem, » Samanyou Garg said during the webinar.
On an e-commerce site, I apply the same principle with a 3-day tight loop. A catalog of 1,200 product sheets can’t afford to lose a citation on a bestseller for three weeks. Every Monday, a script pulls AI snippets on 200 strategic queries. Tuesday, prioritization. Wednesday, deploy fixes. The loop works because it’s mechanical, not heroic.
According to Writesonic’s analysis, the vast majority of links in AI responses direct users to external platforms like Reddit, YouTube, and forums. This chart shows the breakdown.
96% of AI citations point to third-party sites
Only 4% of citations refer to your own domain
Lesson 2: 96% of citations point to pages you don’t own
The number stings. 96% of links in AI responses go to external sources: Reddit, YouTube, forums, news sites. Six months ago it was 80%. The shift from GPT 5.3 to GPT 5.5 accelerated the trend.
Your site is just an island. Without presence on these third-party platforms, your products vanish from conversational responses. Writesonic built an agent that automatically creates a list of cited sites for each prompt where the brand is missing, with author contact info. Result: you establish yourself on the right Reddits, the right YouTube threads, the right forum pages.
For an e-commerce site, the connection is direct. I identify the 30 prompts that generate product recommendations. I spot the Reddit pages, video comparisons, cited forums. Then I publish content there. Not spam links. Expert opinion, product testing, detailed answers. Google’s AI agent cites what’s useful. Be useful where citations are.
Key stat: 96% of AI citations point to sites you don’t control. Your content strategy must cover also external platforms.
Lesson 3: Citation lifespan is measured in days, not months
Writesonic measured the lifespan of 150,000 citations. The average is much shorter than content managers imagine. A citation rarely lasts more than four weeks before being replaced by another source. Language models cycle, ingest new data, and your position can vanish overnight.
« It’s very volatile, because the models are probabilistic by nature, » Garg explained. A simple index refresh can tip your citation to a competitor. For an e-commerce site, a product sheet that loses its citation during Black Friday is a hole of 800 to 1,200 € in revenue in hours.
I observe the same cyclicity with my clients. A page optimized with the DOSE framework holds an average of 17 days in AI snippets. Then the algorithm hunts for freshness. The remedy? A systematic refresh plan, not last-minute fixes. Every two weeks, I update the figures, customer reviews, structured questions and answers. The citation stays. Your competitor waits.
Lesson 4: Don’t put all your eggs in your own site
Before, solid SEO strategy was enough to capture traffic. Today, pure organic traffic is just one piece of the puzzle. 96% of citations are external: betting everything on your domain cuts you off from the majority of AI entry points.
The Writesonic team diversifies across Reddit, YouTube, Medium, targeted guest posts and forum profiles. Their agent tracks citations on 47 different platforms. Not for link building. To be present in the ecosystem of sources language models exploit.
For a merchant site, diversification takes three forms to implement right away. One. Publish video comparisons of your products on YouTube, with structured scripts that answer questions AI reformulates. Two. On specialized forums (Reddit, Quora, Doctissimo for health, HFR for tech), post expert answers containing precise data AI searches for: dimensions, weight, compatibility, delivery time. Three. Pitch a buying guide to a partner media in exchange for a natural citation. Each touchpoint increases your odds of being cited.
Three weeks after deploying these three actions on a catalog of 800 SKUs, AI citations jumped +62%. Without a single change to the main site.
Lesson 5: Without an agent, you’re playing roulette
Monitoring 1,200 keywords, checking 47 platforms, measuring the lifespan of 150,000 citations… by hand, that’s too much for one person. Writesonic built dedicated agents that extract, compare and alert. Humans just prioritize and decide.
You don’t have their infrastructure. You’re looking for a system that runs without you. I build that for my clients. A Google Sheet connected to an API, a Python script that queries SERPs with AI snippet extraction, a Slack alert when a product citation vanishes. Cost: 0 € in extra software. Just engineering.
On an auto accessories client’s shop, the agent detected 19 citation disappearances in one week. Each time, a minor update — an 80-word paragraph, a compatibility table, a review excerpt — was enough to recover the spot in 48 hours. Without this loop, those 19 pages would have stayed in the shadows for three weeks. Estimated loss: 2,300 € gross margin.
Automation isn’t a startup luxury. It’s the only way to keep pace with probabilistic models. As Garg sums it up: « AI search turned SEO into an engineering problem. »
Your next action: a citation audit, not a classic SEO audit
You have the numbers. You have the loop. Now look at your own site. Ask yourself: how many of my top 50 product sheets are cited in an AI response today?
Launch a citation audit instead of another On-Page audit. Take ten queries that drive 80% of your revenue. Open Bard, ChatGPT and Google’s generative expérience. Note which pages are cited. If yours isn’t there, identify the cited source. Is it on a forum? On YouTube? That’s your next playground.
Writesonic’s 35% of AI leads aren’t a fluke. They come from honed engineering. A loop, agents, diversification. The cherry on top: these mechanics are accessible to a structured e-commerce site. I can show you.
Is your site visible in AI responses?
I’ll show you in 45 minutes which pages are cited, where competitors take your spot, and how to build an engineering loop to capture citations.
Book a strategic call — 45 minFrequently Asked Questions
Why does AI Search transform SEO into an engineering problem?
Citations change in days across dozens of platforms. An automatic tracking and update system keeps your spot.
How do I know which sources are cited for my keywords?
I do it manually for my 10 main queries, I note the links. To automate, I use a script that retrieves citations via search APIs or no-code agents.
Is the average citation lifespan really 2 to 4 weeks?
Yes, according to Writesonic’s analysis of 150,000 citations. On very competitive e-commerce markets, I sometimes observe cycles of 10 days only.
Do I really need an AI agent to monitor all this?
At small scale, a tracking file and manual queries are enough. Beyond 50 keywords, automation is necessary to avoid traffic loss.
Is this applicable to an e-commerce site or only SaaS?
Yes, it’s fully applicable. Product sheets, buying guides and comparisons are widely picked up by AI responses. The loop and diversification mechanics work identically.
