From invisible to indispensable: 6.4× AI citations for Northwind Analytics
A B2B analytics platform was losing every AI answer to two competitors. We rebuilt its entity layer, published a quotable corpus, and grew weekly citations 6.4× in 120 days.
- 6.4×
- Weekly AI citations across six engines
- 41%
- Share-of-answer on 60 core buyer prompts
- 18%
- Of trial signups now AI-referred
The situation
Northwind sells a product analytics platform to mid-market data teams. It ranked well in classic search — page one for most category terms — but appeared in just 9% of AI answers for the prompts its buyers actually typed into ChatGPT and Perplexity. Two competitors owned the rest.
The team had already published a large content library. The problem was not volume. It was that almost none of it read like a source an answer engine could extract, and the brand’s entity footprint was inconsistent across the web.
What we found
- No canonical definition. Northwind described itself six different ways across its homepage, LinkedIn, G2, Crunchbase and press coverage. Models were resolving it as three loosely related products.
- Unquotable pages. Articles opened with 300 words of positioning before answering the question in the title.
- No author entities. Content was published under the company byline, with no human expertise attached — one of the strongest retrieval signals in technical categories.
- Blocked retrieval.
robots.txtblocked two major AI crawlers, inherited from a 2021 security review nobody had revisited.
What we built
We ran a 60-prompt panel sampled weekly across ChatGPT, Perplexity, Gemini, Claude, Copilot and AI Overviews. Then we worked in three layers.
Entity layer. One fact sheet, one description, one set of claims — propagated to the site, G2, Crunchbase, LinkedIn and 14 third-party profiles. Organization, Product and Person schema added, with authors linked to real profiles.
Corpus layer. Twenty-two pages rebuilt around extractable answers: a definition page, four comparison pages, three methodology explainers, and a benchmark study with original data. Each opens with a direct answer, then supports it with evidence.
Citation layer. We earned placements in four industry roundups and two analyst briefings, and supplied data to three newsletters that models already retrieve for the category.
The results
Weekly citations grew from 34 to 218 across the panel. Share-of-answer on the 60 core prompts rose from 9% to 41%, passing one competitor outright. AI-referred trial signups reached 18% of total, and — importantly — those trials converted to paid at 1.3× the account average.
The program is now in its second year, with the panel expanded to 150 prompts.