From Zero to 295,000 Search Impressions in 90 Days

In the spring of 2026 we did something most agencies never do: we pointed our entire methodology at our own website, set a baseline of practically zero, and tracked every number in public view. Ninety days later, margen.net had earned over 295,000 Google Search impressions — with no advertising spend and no purchased links. This is exactly how.

The Challenge

Starting from a near-zero baseline is the hardest place to be in search. There is no existing authority to lean on, no backlog of indexed pages, and no entity footprint for Google or AI engines to recognise. Worse, the old site was a slow, heavy WordPress build whose markup made it hard for crawlers to understand what we did or who we were.

We also set ourselves a harder target than most. It was not enough to rank in classic search — we needed to be found and cited by AI answer engines (ChatGPT, Perplexity, Google AI Overviews, Claude) at the same time, because that is the discipline we sell. The bar was dual-channel visibility from a standing start.

The Strategy

We ran a single, layered playbook in deliberate order. Each layer made the next one work harder.

1. A technical foundation that gets out of the way (SEO). We rebuilt the site as a fast, static, clean-markup site. Fast pages with semantic HTML are easier for both Google and AI crawlers to parse. We explicitly welcomed AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended) in robots.txt and published an llms.txt mapping our key pages for large language models. Structured data — Organization, Article, FAQPage and Service schema — went on every relevant template so machines understand what each page is and who published it.

2. Content mapped to real intent, at depth (AEO). This is where the curve came from. We mapped the actual questions our market asks — by intent — and built focused content for each: definitions, comparisons, “how to,” pricing, and sector-specific questions. Every page opens with a direct, 40–60 word answer to its title question before any preamble, because that block is exactly what Google lifts into a featured snippet and what AI engines extract as a citation. FAQ schema on every relevant page captured the question-shaped impressions that plain article content misses. Breadth of genuinely useful coverage — not word count for its own sake — is what turned a handful of impressions into hundreds of thousands.

3. Entity authority and internal structure (GEO). Content makes you eligible; authority gets you chosen — by Google and by the LLMs. We built consistent entity signals (naming, schema, sameAs), cornerstone topic hubs with rings of supporting content linking back to them, and sourced, specific writing (AI engines preferentially cite content with concrete figures and named sources).

4. Fast indexing. A page earns zero impressions until it is indexed. We submitted new and updated URLs via IndexNow the moment they published, so they were picked up in hours rather than weeks.

The Results

• Over 295,000 search impressions earned in 90 days, from a near-zero baseline.
• 1,471 keywords ranking in Google’s top 10 — 400 of them in the top three — and position #1 for our own category term, “margen ai.”
• A growth curve with the classic content-and-entity signature: roughly flat for the first month, then a steep, sustained climb as indexed content and entity signals began to compound.
• A fast, static, schema-complete build that welcomes AI crawlers and ships an llms.txt — engineered to be parsed and cited by both Google and the LLMs.
• Visibility across both channels at once — classic search results and AI answer engines — from a standing start.

Cited by Every Major AI Engine

Impressions prove Google noticed us. The harder test — the one we actually sell — is whether AI engines cite us. They do. Over the same 90 days, all five leading AI search platforms sent margen.net real, engaged visitors: ChatGPT, Google’s Gemini, Claude, Perplexity and Microsoft Copilot.

And these were not glance-and-leave visits. Readers arriving from ChatGPT and Gemini averaged over five minutes on the page; on our AI-citation research specifically, the average visit ran past twelve minutes. AI-referred readers land exactly where the citations point — our statistics and citation-behaviour research, the sourced, answer-first content AI engines prefer to quote. That is what “being the cited source” looks like in the numbers, not just the pitch.

The shape of the curve is the real lesson. Early on, you are seeding pages and establishing trust, and little shows. Once a critical mass of well-structured content is indexed and the entity signals connect, every new page compounds on the authority of the last — and impressions accelerate. The patience in weeks one to four is what pays off in weeks six to twelve.

What Made It Work

1. The foundation came first — speed, clean markup, schema, and crawl access for AI bots, so the content was eligible.
2. Intent was mapped, then answered first — real questions, one focused page each, with the answer in the first 60 words.
3. Entity authority was built deliberately — schema, topic hubs, internal links and sourced content, to become the trusted source, not just a source.
4. Nothing waited in a queue — fast indexing turned new content into impressions in hours.
5. Every page served both channels — built to rank in search and be cited in AI answers.

That is the playbook. We ran it on ourselves first so we could prove it works before we run it for you.