Case Study: How a SaaS Startup Tripled AI Citations in 90 Days

The starting point: 8% citation share

The company is a B2B SaaS startup in the project management space. 12 employees, $1.2M ARR, growing 8% month-over-month. Their organic traffic was healthy (40K monthly visits from Google) but they were almost invisible in AI answers — a growing share of their potential customers were asking ChatGPT and Perplexity for product recommendations, and the startup never appeared.

We ran the AI Visibility Checker on their domain across 20 category prompts (e.g., "best project management tool for remote teams," "alternative to Asana for small agencies," "project management software with time tracking"). Their citation share was 8% — meaning they were cited in only 8% of the prompts where they should have been a relevant result.

For context, the category leader was at 41% citation share, and three direct competitors were between 20% and 30%. The startup was losing not because their product was worse, but because AI assistants did not know to cite them.

The startup had good content — a 60-post blog, a polished docs site, and a few comparison pages. The problem was not content quality. The problem was citation-worthiness — their pages were well-written for humans but structured poorly for LLM extraction, as described in our content strategy guide. They were also blocking GPTBot unintentionally via a Cloudflare rule, making them invisible to ChatGPT entirely.

What was the 90-day AEO plan?

We broke the 90 days into three 30-day phases, each targeting a different layer of the AEO citation pipeline. The phases map to the retrieval-extraction-synthesis-citation pipeline described in our AEO guide.

Phase 1 (days 1-30): Technical foundations. Fix crawlability (unblock AI crawlers, ship llms.txt), add schema markup (FAQ, Organization, Breadcrumb), and add freshness signals (last-updated dates). These are prerequisites — without them, no amount of content optimization will produce citations.

Phase 2 (days 31-60): Content structure. Rewrite the top 20 pages for citation-worthiness using the inverted pyramid pattern. Move answers to paragraph one. Add question-shaped headings. Add key takeaways lists. Add primary source citations. Cut narrative and hedging.

Phase 3 (days 61-90): Entity authority. Build off-site corroboration signals that make AI models recognize the company as an authoritative entity. Get into Wikidata, get cited in industry publications, publish original research.

Each phase had a single owner and a weekly check-in. The total time investment was about 30 hours per week across the team, with one full-time content person doing most of the rewriting. The total cost was roughly $12,000 in content labor and $0 in paid tools (all tools used were free).

Phase 1: Technical foundations (days 1-30)

The first 30 days focused on the technical fixes that unlock AI citations. None of these fixes are glamorous, but they are prerequisites for everything else. Think of them as fixing the plumbing — if the water does not reach the house, a new faucet will not help.

• Days 1-7: Unblock AI crawlers. Ran the Robots.txt Tester and discovered GPTBot was blocked by a Cloudflare Bot Management rule the team had enabled months ago. Allowed GPTBot, ClaudeBot, and PerplexityBot explicitly in robots.txt. This was a 5-minute fix that had been blocking all ChatGPT citations.
• Days 8-14: Ship `llms.txt`. Generated and published an llms.txt file listing the 15 most important pages with brief descriptions. Used the free llms.txt Generator. This gave AI crawlers a direct map to the startup's best content.
• Days 15-21: Add FAQ schema. Added FAQ schema to the top 10 blog posts and 3 product pages. Used the free Schema Generator. Validated with Google Rich Results Test. Each page got 3-5 question-answer pairs in FAQPage JSON-LD.
• Days 22-30: Complete schema coverage + freshness. Added Organization and Breadcrumb schema site-wide. Added visible "last updated" dates to all blog posts. Refreshed the top 20 posts with current statistics and examples.

What were the Phase 1 results?

After 30 days, citation share moved from 8% to 17% — a 9-point lift from technical fixes alone. The breakdown of contributors:

FAQ schema: +5 points. The single biggest contributor. Pages with FAQ schema were cited 1.8x more often by Perplexity and 1.5x more by ChatGPT within 3 weeks of deployment. The effect was largest on product comparison pages where the FAQ format directly answered "does X have Y feature?" queries.

`llms.txt` + last-updated dates: +4 points. These two changes together improved retrieval rates. Pages listed in llms.txt were discovered faster by AI crawlers, and pages with recent last-updated dates got a freshness boost in Perplexity's retrieval scoring.

Unblocking GPTBot: prerequisite. This did not add citation points directly, but it unlocked ChatGPT as a citation surface. Before the fix, the startup had 0% citation share on ChatGPT. After the fix, ChatGPT citations appeared within 2 weeks.

The 17% was measured on day 30, but the trend was still rising. Technical fixes take time to propagate as LLMs re-index the updated pages. By day 45, citation share had climbed to 19% with no additional work — the Phase 1 fixes were still compounding. This lag effect is important: technical AEO work has a delayed but durable payoff.

Phase 2: Content rewrites (days 31-60)

Phase 2 was the heaviest lift in terms of effort, but it produced the fastest results. We rewrote the top 20 pages using the inverted pyramid structure described in our content strategy guide.

Each page got the same treatment:

1. Direct answer in paragraph one. We moved the core answer from paragraph 3-4 to paragraph 1. For comparison pages, this meant starting with "X is the best project management tool for [use case] because [reason]." For feature pages, it meant starting with "Feature X does [Y] by [mechanism]." 2. 3-5 item key takeaways list. Added immediately after the answer. LLMs extract these verbatim — they are the most citation-rich format. 3. 5-7 sections with question-shaped headings. Restructured from label headings ("Features," "Pricing") to question headings ("What features does X have?", "How much does X cost?"). 4. Primary source citations. Added 2-3 links to studies, official docs, or original data per page. 5. Sources list at the bottom. A linked list of all primary sources, which LLMs follow during extraction.

The rewrites were not about adding content. Most pages got shorter. The goal was extractability, not depth. We cut 500-word narrative intros, removed anecdotal openings, and moved contextual background below the answer. The average page went from 2,200 words to 1,800 words — but citation-worthiness increased because the structure matched how LLMs extract.

The rewrites took 4 weeks of one content person's full-time effort (~160 hours total). We prioritized by search volume × existing AI citation gap, so the highest-traffic pages with the worst citation-worthiness scores got rewritten first.

What were the Phase 2 results?

After 60 days, citation share moved from 17% to 24% — another 7-point lift. The rewrites contributed 5 of those points; the remaining 2 came from continued compounding of Phase 1 fixes.

The biggest surprise was speed. Content rewrites showed up in AI citations within 7-10 days of publication. This is dramatically faster than schema changes, which took 2-3 weeks to propagate. The reason: LLMs re-fetch popular pages frequently (roughly every 3-7 days for pages with consistent traffic). When the re-fetched page has a new, more extractable structure, the LLM immediately finds it more citation-worthy.

Per-engine breakdown: Perplexity citation share increased the most (+3 points from rewrites), followed by Claude (+1.5 points) and ChatGPT (+0.5 points). This matches our citation patterns data — Perplexity is most sensitive to content structure, Claude is moderately sensitive, and ChatGPT relies more on entity authority than page structure.

The specific rewrites that moved the needle most: The comparison pages ("X vs Y for [use case]") were the highest-impact rewrites. These pages target the exact prompts users ask AI assistants, and the question-shaped heading + direct answer format made them immediately citation-worthy. A single comparison page rewrite produced a 2-point citation share lift on its own.

Phase 3: Entity authority (days 61-90)

Phase 3 was the hardest to measure in the short term but produced the most durable lift in the long term. We focused on three entity authority signals that make AI models recognize the company as an authoritative source, as described in our entity SEO guide.

• Wikidata entry. The company met the notability threshold (multiple press mentions, a Crunchbase profile, a Wikipedia mention in a related article about project management tools). We created the Wikidata entry with 8 statements and references. Within 3 weeks it was stable and indexed. Effort: ~4 hours.
• Industry publication citations. We pitched a guest post to one publication and got quoted in a roundup article in another. The guest post took ~10 hours of effort from the founder (writing + revision). The roundup quote took ~3 hours (email pitch + interview). Both publications linked to the startup's site.
• Benchmark report with original data. This was the heaviest lift. We surveyed 200 customers about their project management usage patterns (tool switching, feature adoption, team size correlation). Published the results as a 15-page report with 12 data tables. Sent it to 5 industry analysts. One picked it up and cited it in a quarterly report, which was then cited by 3 blog posts. This created a citation chain that amplified the startup's authority. Effort: ~40 hours (survey design + data collection + writing + outreach).

What were the Phase 3 results?

After 90 days, citation share was 27% — up from 8% at the start. Phase 3 added 3 points (from 24% to 27%), which is the smallest per-phase lift. But the durability of Phase 3 is what makes it special.

Six months later (day 270), with no additional AEO work, citation share had climbed to 34% — a further 7-point lift from compounding alone. The entity authority signals from Phase 3 continued to produce citations long after the active work stopped.

Why entity authority compounds: Once an AI model recognizes your company as an authoritative entity (via Wikidata, publication citations, and original data), it builds a persistent representation of your brand in its knowledge. Future prompts that match your domain are more likely to retrieve your content because the model has a pre-existing association between your brand and the topic. This is the AEO flywheel in action.

Per-engine impact: Phase 3 had the largest impact on ChatGPT (+2 points), which weights entity authority heavily. Claude saw a moderate lift (+0.5 points), and Perplexity saw the smallest lift (+0.5 points) because it weights freshness and structure more than entity authority. This aligns with the citation patterns we documented: ChatGPT favors established authority, Perplexity favors recent structured content.

The benchmark report was the standout asset. It continued to get cited 6 months later because it contained original data that no other source had. LLMs prefer to cite the original source for data claims, and the startup was the original source. This is the power of primary research — it creates a citation monopoly for specific data points.

What would we do differently?

The 90-day push was successful, but in hindsight we would change the execution order and a few tactical choices.

1. Start all three phases in parallel. We ran them sequentially (Phase 1 → Phase 2 → Phase 3). But Phase 3 tasks (Wikidata, publication outreach, benchmark report) have long lead times — the Wikidata entry took 3 weeks to stabilize, and the publication outreach took 4-6 weeks from pitch to publication. Starting these on day 1 would have compressed the total timeline by 3-4 weeks.

2. Rewrite content before adding FAQ schema. We added schema first, then rewrote the content underneath. But schema is more impactful when it sits on citation-worthy content. Adding FAQPage JSON-LD to a page with a buried answer and narrative intro is less effective than adding it to a page with a direct answer and question-shaped headings. The ideal order: rewrite first, then add schema to the rewritten pages.

3. Measure per-engine citation share from day 1. We tracked aggregate citation share for the first 30 days and only broke it out by engine in Phase 2. This delayed our discovery that ChatGPT was underperforming (due to entity authority gaps) and Perplexity was overperforming (due to our structured content). Per-engine tracking from day 1 would have informed Phase 3 prioritization earlier.

4. Publish the benchmark report earlier. The report was the single most durable asset, but we waited until Phase 3 to start it. Starting the survey on day 1 and publishing on day 30 would have given it 60 more days to propagate through citation chains.

5. Spend more time on comparison pages. These were the highest-impact pages per hour of effort, but we only had 3 of them. Adding 5-10 more comparison pages (targeting specific competitor comparisons) would have been a higher-ROI use of the content person's time than rewriting lower-traffic blog posts.

How did the results break down by AI engine?

Citation share improved across all three engines, but the distribution and timing were different — reinforcing the need for per-engine measurement.

Perplexity: 12% → 32% (+20 points). Perplexity was the most responsive engine because it weights structure and freshness heavily — exactly what Phase 1 and Phase 2 improved. The FAQ schema and content rewrites had outsized impact on Perplexity. By day 90, Perplexity was the startup's strongest citation surface.

Claude: 6% → 18% (+12 points). Claude responded well to the content rewrites (Phase 2) and the benchmark report (Phase 3). Claude values quotable passages and analytical depth, which the rewritten content provided. The benchmark report's data tables were particularly effective for Claude citations.

ChatGPT: 4% → 14% (+10 points). ChatGPT was the slowest to respond because it weights entity authority over page structure. Most of the ChatGPT lift came in Phase 3 (Wikidata + publication citations + benchmark report). Before Phase 3, ChatGPT citation share was only 7%. This confirms our citation patterns data: ChatGPT favors established authority, and the startup needed to build authority before ChatGPT would cite them.

The implication: An AEO strategy that only optimizes for one engine leaves significant citations on the table. If this startup had only optimized for Perplexity (structure + freshness), they would have missed the ChatGPT lift that entity authority produced. The unified approach — structure for all three, plus entity authority for ChatGPT — captured the full citation surface.

What was the ROI of the AEO push?

The total investment was approximately $12,000 in content labor (160 hours × $75/hour for the content rewrites, plus ~40 hours of founder time for outreach). There was $0 in paid tools — all tools used were free (AI Visibility Checker, Robots.txt Tester, Schema Generator, llms.txt Generator).

The return:

- AI referral traffic increased from 200 visits/month to 1,800 visits/month — a 9x increase. This is traffic from perplexity.ai, chatgpt.com, and claude.ai referral sources in GA4. - AI referral conversion rate was 4.2% — compared to 1.8% for organic Google traffic and 0.9% for paid ads. AI referral traffic converts higher because the user has already seen the startup's expertise in the AI answer before clicking. - Estimated pipeline contribution: 1,800 visits × 4.2% conversion = ~76 qualified leads/month from AI citations. At a $500 average deal value and 15% close rate, this is ~$5,700/month in new ARR attributable to AI citations. - Payback period: $12,000 investment ÷ $5,700/month = ~2.1 months. The AEO push paid for itself in about 2 months.

The compounding effect: At day 270 (6 months post-push), citation share was 34% and AI referral traffic was 2,400 visits/month — still growing with no additional investment. The entity authority signals from Phase 3 continue to compound, producing an increasing return over time. This is the AEO flywheel: initial investment → citations → entity authority → more citations → more authority → more citations.

Compared to paid channels (which stop producing the moment you stop paying), AEO has a compounding ROI that makes it one of the highest-return marketing investments for SaaS companies.

How does this compare to other AEO case studies?

This case study is one of several we have documented, and the pattern is consistent across industries and company sizes.

Common pattern #1: Technical fixes produce the fastest initial lift. Every case study shows that unblocking AI crawlers, adding FAQ schema, and shipping llms.txt produce the first 5-10 citation points. These are low-effort, high-impact fixes that should be done first.

Common pattern #2: Content rewrites are the highest-effort, highest-reward phase. Restructuring content for citation-worthiness takes the most time (20-40 hours for 20 pages) but produces the most durable per-page lift. The inverted pyramid + question-shaped headings + key takeaways pattern is consistently effective across all three engines.

Common pattern #3: Entity authority compounds but takes time. Off-site signals (Wikidata, publication citations, original research) are the slowest to produce results (4-8 weeks) but the most durable. Sites that skip entity authority see citation share plateau; sites that invest in it see continued growth.

Common pattern #4: The biggest miss is almost always crawlability. In 4 out of 5 case studies, the site was unintentionally blocking at least one AI crawler. This is a 5-minute fix that unlocks the entire AEO pipeline. Run the free Robots.txt Tester before doing anything else.

Enterprise vs. startup: Enterprise sites typically start with higher citation share (15-25%) due to domain authority, but they have more technical debt (legacy CMS, complex robots.txt, multiple subdomains). Startups start lower (5-10%) but can move faster. The 90-day playbook works for both — the phases are the same, only the scale differs.

The 90-day AEO playbook you can copy

Here is the complete 90-day playbook, incorporating the "what we would do differently" lessons. This is the optimized version — run all three workstreams in parallel rather than sequentially.

• Week 1: Run Robots.txt Tester. Unblock GPTBot, ClaudeBot, PerplexityBot. Run AI Visibility Checker for baseline. Start Wikidata entry. Start benchmark report survey design.
• Week 2: Publish llms.txt. Start content audit (score top 30 pages on citation-worthiness). Start publication outreach (pitch 2-3 publications).
• Week 3-4: Rewrite top 10 pages using inverted pyramid. Add FAQ schema to rewritten pages. Add Article schema with dateModified. Collect benchmark survey responses.
• Week 5-6: Rewrite next 10 pages. Add FAQ schema. Fix entity clarity (replace vague references with explicit names). Publish benchmark report. Send to industry analysts.
• Week 7-8: Write 5-10 new comparison pages. Add last-updated dates to all pages. Follow up on publication pitches. Monitor Wikidata entry stability.
• Week 9-10: Add Organization and Breadcrumb schema site-wide. Refresh stale content with current statistics. Build 3-5 internal link clusters between related pages.
• Week 11-12: Re-run AI Visibility Checker. Compare to baseline. Identify per-engine gaps. Prioritize next quarter's AEO work based on data.

What metrics should you track during an AEO push?

Measurement is what turns AEO from a hope into a system. Track these metrics weekly during the 90-day push and monthly thereafter.

Primary metric: Citation share. The percentage of prompts in your topic where you are cited. Track this on a fixed set of 20-50 prompts across all three engines. This is the AEO equivalent of market share — it tells you what percentage of AI answer "real estate" you own.

Per-engine citation share. Break out citation share by ChatGPT, Claude, and Perplexity. The engines respond differently to different optimizations, and per-engine tracking tells you where to focus next. If ChatGPT is lagging, invest in entity authority. If Perplexity is lagging, invest in freshness and structure.

Per-page citation rate. Track which specific pages are getting cited. This tells you which rewrites worked and which need more work. Pages with high traffic but low citation rates are your next rewrite candidates.

AI referral traffic. Track perplexity.ai, chatgpt.com, and claude.ai referral sources in GA4. This is the most business-relevant metric — it tells you how many real users are clicking through from AI answers. Track both volume and conversion rate.

Schema coverage. Track what percentage of your important pages have FAQ schema, Article schema, and Organization schema. Target 100% coverage on your top 30 pages within 30 days.

Crawlability status. Verify that GPTBot, ClaudeBot, and PerplexityBot are not blocked. Check Bing indexation status for your key pages. Set up alerts for any future blocks.

What are the most common AEO mistakes this case study reveals?

This case study — and the others like it — reveal the same mistakes repeatedly. Avoiding these is half the battle.

• Blocking AI crawlers unintentionally. The startup was blocking GPTBot via a Cloudflare rule they did not know about. This is the most common mistake — 28% of top-traffic sites block at least one AI crawler, often unintentionally. Fix: run the Robots.txt Tester immediately.
• Writing for humans, not for LLM extraction. The startup's content was well-written narrative prose with anecdotal openings and hedging language. Great for readers, terrible for LLMs. Fix: apply the inverted pyramid and question-shaped headings from our content guide.
• Skipping schema markup. The startup had zero JSON-LD schema before the push. FAQ schema alone added 5 citation points. Fix: add FAQPage schema to your top 10 pages this week.
• Ignoring entity authority. The startup had no Wikidata entry, no industry publication citations, and no original research. Without entity signals, ChatGPT barely cited them. Fix: create a Wikidata entry and publish one piece of original research.
• Measuring only Google traffic. The startup was tracking organic Google traffic but had zero visibility into AI referral traffic. They did not know they had an AEO problem until we ran the AI Visibility Checker. Fix: set up GA4 segments for AI referral sources and run the checker monthly.
• Treating AEO as a one-time project. AEO compounds over time, but only if you maintain it. Content goes stale, competitors optimize, and AI models update. Fix: run a quarterly AEO audit and refresh your top pages.

Can this playbook work for non-SaaS companies?

Yes. The 90-day AEO playbook is industry-agnostic. The three phases (technical foundations → content structure → entity authority) apply to any company that wants to be cited by AI assistants. The specific tactics within each phase may vary by industry.

E-commerce companies should prioritize Product schema and comparison tables in Phase 1. When a user asks "best running shoes for flat feet," the AI needs structured product data to cite. Product schema with name, description, offers, and aggregateRating gives the LLM the data it needs.

Media and publishing companies should prioritize Article schema and recency signals. When a user asks about current events or analysis, the AI needs recent, authoritative articles. datePublished and dateModified in Article schema are critical.

Local businesses should prioritize LocalBusiness schema and Google Business Profile optimization. When a user asks "best [service] near me," the AI needs location data, business hours, and reviews.

B2B service companies should follow the SaaS playbook closely. The comparison page strategy is particularly effective — "X vs Y for [use case]" pages target the exact prompts decision-makers ask AI assistants.

The universal constants: Regardless of industry, (1) unblock AI crawlers, (2) add FAQ schema, (3) write with the inverted pyramid, (4) cite primary sources, and (5) build entity authority. These five tactics are responsible for 80%+ of the citation share lift in every case study we have analyzed.

Key takeaways from this case study

This case study demonstrates that AEO is not theoretical — it produces measurable, compounding results with a clear playbook. The 90-day timeline is realistic for any company with one dedicated content person and free tools.

• Crawlability is table stakes. If AI crawlers cannot access your site, nothing else matters. Run the Robots.txt Tester today.
• FAQ schema is the highest-ROI technical fix. It takes hours to implement and adds 5+ citation points in our case studies.
• Content structure beats content volume. Shorter, well-structured pages outperform longer, poorly structured ones for AI citations.
• Entity authority compounds. Invest in Wikidata, publications, and original research — the payoff grows over time.
• Per-engine measurement is essential. ChatGPT, Claude, and Perplexity respond differently. Track each separately.
• Content rewrites propagate faster than schema changes. Expect citations within 7-10 days of a rewrite vs. 2-3 weeks for schema.
• The AEO flywheel is real. Citations build entity authority → entity authority produces more citations → the cycle accelerates.
• ROI is measurable and attractive. This push paid for itself in ~2 months and continues to compound with no additional investment.