AI search now drives 35% of Writesonic leads. Here's the system.

Twelve months ago, 2.5% of Writesonic's inbound leads came from AI search. By March 2025, that figure had reached 35%. Search Engine Journal covered the system behind the shift in a webinar led by Samanyou Garg, Writesonic's founder and CEO: a repeatable, agent-driven workflow that monitors citation behaviour across every major AI search platform and closes the gaps.

The headline number deserves scrutiny before admiration. Writesonic sells AI writing software, which means its audience already lives inside the AI tool stack. Its citation gains are probably easier to achieve than they would be for, say, a multilateral development bank or an industrial conglomerate whose subject matter is denser and whose publication cadence is slower. The 2.5% to 35% move is real, but treat it as a ceiling for a native AI brand, not a baseline for everyone.

What matters more is the mechanism.

The six-stage loop and what it reveals about the new SEO

Garg's core claim, quoted by Search Engine Journal, is pointed: "AI search didn't necessarily kill SEO, but it turned it into an engineering problem." The six-stage loop his team runs on every published page is the practical expression of that claim. Agents surface movement across platforms; humans prioritise and act. The cycle repeats continuously, not quarterly.

That rhythm represents a structural break from conventional SEO. Traditional search optimisation assumed a relatively stable index and a human-readable signal of performance (rankings). AI citation patterns are more volatile, more opaque, and split across ChatGPT, Perplexity, Google AI Overviews, Copilot, and others simultaneously. A page that wins citations in one model may be invisible in another because training data, retrieval architecture, and recency weighting differ. A single audit cadence cannot cover that surface area. Automation is not optional; it is the only way to maintain visibility at the pace these platforms change.

The second lesson from the Writesonic case is about pages you do not own. The workflow explicitly targets third-party citations: the Reddit threads, industry publications, analyst summaries, and forum answers that AI models prefer when they trust the brand's owned content less. For B2B brands, this is where most of the citation gap actually lives. LLMs trained on the open web will cite a Gartner brief, a UN agency report, or an IEEE standard before they cite a brand's product page. The implication is that earned presence on authoritative third-party sources is a direct input to AI visibility, not a soft benefit of good PR.

Who this actually affects

For financial services brands, the earned-media angle is particularly acute. Models treat regulatory documents, analyst reports, and financial press citations as high-trust sources. A firm that appears in Financial Times coverage, BIS working papers, or IMF publications is far more likely to surface in an AI-generated answer on macro risk or capital adequacy than one relying on owned content alone. The Writesonic workflow of "winning citations on pages you don't own" maps directly onto that reality.

Multilaterals and UN system organisations face a different version of the same problem. They are, in principle, among the most credible sources a language model could cite. The World Bank, UNDRR, and CGAP produce primary data that practitioners and policymakers actually use. But citation does not follow credibility automatically; it follows structured, retrievable content. If a key findings page buries its statistics inside a PDF rather than marking them up in HTML, the model may cite a secondary source that extracted and republished the same number. The engineering problem Garg describes applies here too: publication is not distribution, and distribution is not citation.

For major industrial groups, the challenge is volume and velocity. Holcim or an equivalent publishes sustainability reports, technical specifications, and regulatory filings across dozens of jurisdictions. Monitoring which of those documents are being cited, in which models, for which query types, and at what frequency requires exactly the kind of agent infrastructure Garg describes. A human team running quarterly reviews will always be behind.

The counter-argument worth taking seriously

There is a reasonable objection to the Writesonic framing. Treating AI visibility as an engineering problem risks optimising for current model behaviour rather than for the underlying quality of content and reputation that drives it. Models update. Citation patterns shift. A workflow tuned to win citations in Perplexity's March 2025 configuration may be misaligned by September. The durable answer is not to game each model's current preferences but to build the kind of authoritative, well-structured, widely-cited presence that any reasonable retrieval system would surface.

Garg's system is not incompatible with that view. The six-stage loop and the third-party citation strategy are consistent with genuine quality investment. The risk is if the engineering framing crowds out the editorial one: brands that optimise for citation patterns without the underlying substance will find the advantage temporary.

The 35% figure is a destination. The more transferable insight is that Writesonic got there by treating AI search as a discipline with its own feedback loops, not as a channel that runs itself.