I spent six weeks watching ChatGPT pick our competitors. Here's the playbook that fixed it.

In April I pulled an Ahrefs report on our fractional CMO category. We ranked top-three for almost everything we cared about. I was pleased. I was also wrong.

Six weeks later I ran the same questions through ChatGPT and Claude. Neither one cited us. Both cited three competitors I had not heard of. Two of those competitors did not rank in the top ten on Google for the same query.

That was the moment I understood that the rules had changed.

The data backs up the discomfort.

Share of LLM (SoL or SoLLM) is the percentage of times a brand gets cited, recommended, or mentioned by AI models when buyers ask category questions. It is not a single platform metric, because buyers do not use a single platform. ChatGPT serves 87% of current AI referrals. Perplexity is the research destination for analysts. Claude synthesizes. Gemini is rising fast. Each one cites differently.

You measure SoL by running a fixed set of buyer queries across all of them and counting citation share. The process takes a few hours per quarter. The signal is durable.

The deepest insight from the AEO research is that AI engines re-cite sources they have already learned to trust. ChatGPT shows recursive authority, sources cited by other cited sources rank disproportionately well. Once you are in the citation graph, every new citation reinforces every prior one.

That dynamic creates a moat. Brands that establish citation share in the next 24 months will compound an advantage that late entrants cannot easily climb. The brands that wait until late 2027 to "do AEO" will be optimizing against a graph that already has its trusted sources.

This is why our team is treating Share of LLM as the most important marketing investment we will make in the next 18 months. Not just for ourselves but for every AI startup we operate.

The Princeton GEO study and 2026 AEO weighting research point at five levers that actually move citation share.

If Gartner's projection holds, traditional search volume drops 25% by the end of 2026. If LLM-referred conversion remains 5x organic, the relative leverage of Share of LLM grows roughly 6x in two years. Even the conservative read is that SoL becomes the dominant visibility metric by Q2 2027.

The companies that act before the citation moat closes will compound through 2030. The companies that wait will pay the late-mover tax.

This blog post is one of the bets. We will keep running the SoL audit on our own pages, on Lynqo, on every IG portfolio company, and we will publish the deltas quarterly.

In late March 2026, I ran a query baseline for one of our portfolio companies. The query was simple: "best fractional CMO for AI startups." I ran it on ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews. Three runs each.

Our portfolio company got cited zero times out of fifteen runs. A competitor I respected (but who I knew operated at a much smaller scale than us) got cited eleven times out of fifteen. The competitor's domain authority was lower than ours, their content output was lower, and their team was smaller. They just had more citation density in the surfaces that matter now.

That week was the day Share of LLM stopped being an interesting metric and became the metric we organize the program around.

The next 60 days produced the playbook we now run for every portfolio company starting from a citation deficit. Five moves, all of them shippable in a single sprint each.

The first version of our citation-magnet content overshot on length and undershot on answer-first formatting. We wrote 3,000-word pieces with the answer buried in section 4. AI surfaces extracted the wrong sentences and our citation share moved slowly.

The second pass inverted the structure. Answer in the first 40-60 words. Supporting depth below. Statistical anchors integrated into the answer block, not relegated to a stat band the model ignored. Citation share moved within 14 days of the rewrite.

The lesson: AI surfaces don't reward length. They reward extractability. A 1,500-word piece with an answer-first lead and FAQ schema will out-cite a 3,500-word piece without that structure every time.

Three pieces of advice for anyone starting the SoL journey from a deficit.