How a SaaS Blog Tripled Its AI Citations in 90 Days

Your blog posts are ranking on Google, but when someone asks ChatGPT for a project management tool recommendation, your brand never shows up. A mid-sized SaaS company faced exactly this gap, and closed it in a single quarter.

Background

A mid-sized SaaS company in the project management space had built a respectable content engine. Over three years, its marketing team published 240 blog posts, earned steady organic traffic, and ranked on the first page for dozens of commercial keywords, measured the way most content groups do: keyword rankings, organic sessions, and backlinks.

Then the referral data started shifting. More prospects arrived through typed-in URLs and vague "heard about you from a friend" notes in sales calls. When the team asked how those buyers first encountered the product, a pattern emerged: people were asking ChatGPT, Perplexity, and Google's AI Overviews for tool suggestions, and the AI engines were naming competitors instead.

The company was invisible inside the machines that were increasingly shaping purchase decisions. That realization reframed the team's entire content strategy.

How the team found the leak

Nobody noticed because no one was looking. The turning point was a sales operations analyst who began tagging every inbound lead with a free-text "how did you first hear about us" note. Over one quarter, a recurring theme emerged: prospects describing an assistant's recommendation, often followed by "but I wanted to compare you against them anyway."

That note describes a specific failure mode. The brand was not being rejected, it was being excluded. The team validated the hypothesis with a manual audit of twenty buyer questions across three engines, and the baseline was low.

Challenge

The core problem was not traffic. It was citation. The blog performed well in traditional search but was almost never quoted by generative engines. Three issues stood out.

First, the posts were optimized for keyword matching, not for the way language models extract and recombine information. They opened with long introductory throat-clearing, buried the answer several paragraphs down, and used conversational hedging that gave models little to grab onto.

Second, the site lacked structured signals that help AI systems understand entity relationships: no consistent schema markup, no clear description of what the product was, and no internal linking that established topical authority.

Third, the publishing cadence was inconsistent and the topics scattered. One week brought a pricing explainer, the next a hiring tip, the next a generic trend piece. Models learn topical authority through concentration, and this blog spread itself thin.

The team had ninety days to change that before the next board meeting, where growth metrics would be scrutinized.

The team tested the cheapest hypothesis first. They republished two high-traffic posts with a tightened opening paragraph and nothing else changed. Traffic held, rankings held, citations did not move either. That null result ruled out the easy fix: one isolated edit was never going to change how a model read a three-year archive of 240 posts. The problem was systemic, so the team planned a rewrite program.

What They Did

The team ran a focused GEO program rather than a content volume push. The work broke into five moves.

1. Rewrote the top forty posts for answer extraction. Each post now opened with a direct, self-contained answer to the question it targeted. The team stripped introductory fluff, stated the core claim in the first two sentences, used short declarative sentences a model could lift verbatim, and added a one-paragraph summary block near the top.

The rewrite followed a repeatable template rather than individual judgment. Each target question became the page's opening line, followed by a two-to-three sentence direct answer, then supporting detail under descriptive subheadings, then a numbered list that could be lifted item by item. The template was written once and shared with a freelance editor.

2. Built entity clarity with schema and consistent definitions. They added Organization, Product, and Article schema across the site, and every post defined the product the same way: a project management platform for remote software teams. Repeating one clear description helped models associate the brand with the category.

3. Concentrated topical coverage. The team committed to a cluster around project management for remote teams, publishing twelve deep posts on subtopics within that theme. Internal links connected them into a tight web, signaling authority on one subject rather than shallow coverage of many.

4. Earned mention-worthy data. They published one original benchmark report with proprietary usage numbers from their own customer base. Models favor content with specific, citable statistics, and a self-reported benchmark gave them a quotable asset no competitor had.

5. Monitored citations directly. Rather than guessing, the team asked the same twenty questions of ChatGPT, Perplexity, and Google AI Overviews every week and recorded whether the brand appeared, creating a citation rate they could track.

How the work was staffed and sequenced

Two people carried the work, neither a full-time hire. The content lead owned the template, the priority ordering, and final review; a freelance editor handled the mechanical rewriting; the head of growth ran the weekly citation check.

The order mattered more than the volume. The team rewrote the forty highest-traffic posts first, because those already had authority and were most likely in a model's source pool. Schema came next, as a single template-wide change. The topical cluster launched only after the first rewrites shipped, so new posts inherited the new structure.

Results

After ninety days, the citation rate across the twenty test questions moved from 8 percent to 27 percent. On the specific question "what is the best project management tool for remote teams," the brand appeared in Perplexity's answer for the first time, and in Google AI Overviews within six weeks of the cluster launch.

Qualitative shifts mattered as much as the numbers. Sales reps reported that more inbound prospects mentioned "an AI recommended you." The benchmark report became the single most-cited asset, appearing in answers to data-specific questions and driving referral traffic traditional analytics had not previously captured.

The team did not abandon traditional SEO. Organic sessions held flat, which the team counted as a win given that AI Overviews were cannibalizing clicks across the category anyway.### How the result was measured

The headline figure comes from the team's own tracking sheet, not a third-party tool, and the team was explicit about that limitation internally. Each Monday the twenty questions were run through each engine in a clean session, and a response counted as a citation if the brand was named in the answer or appeared in the source list. The questions stayed fixed for the full ninety days, because a changing set makes comparison meaningless.

Sampling was intentionally modest, and the team accepted that one week could move several points in either direction without meaning anything. The useful reading was the quarter-long trend, reinforced by a monthly count of sales notes mentioning an assistant and a manual record of which URL was cited. That second check was the most actionable output, because it showed pages with a comparison table or a numbered checklist cited far more often.

Lessons

Concentration beats scatter. A tight topical cluster taught models what the brand stood for far faster than a broad publishing stream.

Answer first, elaborate second. The single most effective change was moving the direct answer to the top of every post. Models reward pages that state claims plainly.

Measure what matters. Tracking citation rate directly, rather than inferring it from traffic, gave the team a feedback loop traditional analytics could not provide.

Generative engine optimization is not a replacement for SEO. It is a layer on top of it, for a world where the answer, not the link, is the product.

A checklist you can reuse

• Lock a fixed question set before you change anything, and keep it fixed long enough to read a trend.
• Rewrite the pages that already have authority, not the ones you wish you had.
• Give every page one question and one direct answer, placed at the top.
• Make each list item meaningful when read alone, out of context.
• Use one identical product description everywhere, including the schema.
• Publish something only you can produce, ideally containing your own numbers.
• Record which URL gets cited, and let that guide the next round of edits.

Frequently Asked Questions

How do you measure AI citations without special tools?

You can run a manual citation audit by asking the same set of target questions to ChatGPT, Perplexity, and Google AI Overviews on a fixed schedule, then recording whether your brand appears. Keep the question set unchanged for the whole measurement period, and record which URL was cited as well as whether the brand appeared. A spreadsheet with weekly rows is enough to spot trends.

Should I rewrite old posts or publish new ones for GEO?

Both, but prioritize rewriting. Your highest-traffic and highest-intent existing posts already have authority and backlinks, so restructuring them for answer extraction delivers faster citation gains than starting from zero. Publish new cluster content to build topical depth around the subjects you most want to own, and link it into the pages you rewrote.

How long before AI citations improve after changes?

Most teams see movement within four to eight weeks, because generative engines re-crawl and recompute answers on a rolling basis. The ninety-day window in this case study captured a full quarter of compounding change, which is why the result looked dramatic by day ninety rather than day thirty. Expect noisy swings in between, and judge the work on the trend.