From missing to the first answer

The problem

Northbeam had the search presence you would expect from an established CRM. It ranked on page one of Google for most of its category terms, and the marketing team had a regular publishing schedule. But that visibility did not carry over when a buyer asked an AI assistant which product to choose.

Across the prompts we tracked, Northbeam appeared in just 4% of answers. When it did appear, the description often belonged to an older version of the product: a retired free plan, a missing Gmail integration, or pricing from 2024. A buyer could form an opinion before ever reaching the current website.

The brief was to make Northbeam part of the shortlist for the customers it actually served. More mentions would only be useful if those mentions described the product accurately.

Establishing a useful baseline

We agreed 218 buyer prompts with the sales team and tracked them daily across ChatGPT, Perplexity, Gemini, Google AI Overviews and Copilot. The list covered category recommendations, competitor comparisons and factual questions about the product.

We separated being named from being cited, and logged incorrect facts as their own issue. That distinction mattered: a recommendation containing an old price was not a clean win. Keeping the prompt list stable also meant we could compare later answers with the starting point.

What we did

Weeks 1–2. We turned the baseline into a working queue. Each issue had a buyer question, an observed answer, a likely source and a person responsible for correcting it. This helped the Northbeam team see why a particular page or directory listing deserved attention.

Weeks 3–6. We rewrote the pricing and integration pages and three comparison pages. The new copy put product facts close to the questions buyers asked: what was included, which systems connected, and when Northbeam was the appropriate choice. We also corrected product facts in 14 directories so the website was not contradicted by its surrounding listings.

Months 2–9. We worked on the sources already appearing in the tracked answers. Over the engagement, that meant 23 placements in buyer guides, review sites and relevant Reddit threads. The aim was to contribute accurate, useful information in places buyers and answer engines were already consulting.

Every Friday, the changelog connected completed work to the questions it was intended to affect. If an answer stayed wrong, the issue stayed open rather than disappearing behind a delivered-content count.

What moved

Share of answer increased from 4% to 38% over nine months. Among answers that mentioned Northbeam, it was named first in 61%. Wrong pricing dropped out of ChatGPT answers within five weeks.

The engines did not move together. Perplexity responded fastest in this engagement, while Copilot moved least. We kept those differences visible instead of allowing the blended result to suggest uniform progress.

These are visibility results across a defined prompt set. They do not establish how many buyers saw an answer or how much revenue a particular mention produced.

What we’d do differently

We would start the community work sooner. Threads we joined in month four began appearing in citations within weeks. Bringing that work forward would have tested an important source category earlier, although we cannot know precisely how the earlier timing would have changed the final result.

For a team facing the same problem, the first step is a dependable baseline and a short list of wrong facts. Correct the product story before spending more time spreading it.