We Ran 1,000 Queries on Perplexity. Your Listicle Is Helping Your Competitors.

We ran 100 B2B software queries through Perplexity’s Sonar API, 10 times each, across 10 verticals. 1,000 total results. We tracked every citation, brand mention, and source attribution to answer a simple question: when Perplexity cites your content, does it actually recommend you?

Often, it does not. 25.7% of the time Perplexity used a listicle as a source, it recommended competitors ahead of the content owner. We call this a backfire. Your content gets cited. Your competitors get recommended.

But the data also reveals something the backfire headline obscures: smaller brands are already beating giants on high-value queries, and 25% of buyer queries have no stable leader at all. The opportunity is wide open for companies that build the right content.

The Data

100 queries spanning CRM, marketing, security, finance, HR, project management, dev tools, data infrastructure, AI tools, and customer success. Each query was run 10 times to account for non-deterministic responses. Every result was logged with citations, brand mentions, brand positions, source types, and citation ownership.

Size Does Not Determine Who Wins

The conventional assumption is that AI engines favor the biggest brands. Salesforce, Microsoft, and HubSpot have massive content libraries, dominant G2 profiles, and decades of backlinks. If GEO worked like SEO, they would own every query.

They do not. Giants hold a stable #1 position on 65 of 100 queries. But non-giants hold #1 on 10 queries with 70%+ consistency. And 25 queries have no stable #1 at all.

Apollo is a fraction of ZoomInfo’s size. It holds #1 on the pricing comparison query in 8 out of 10 runs. Vercel is smaller than Cloudflare by every traditional metric. It holds #1 on the deployment platform comparison in 8 out of 10 runs. Asana beats Monday.com and Jira in 9 out of 10 runs on the category discovery query.

The pattern is not brand size. It is category ownership through structured content. Apollo wins on the pricing query because its content is built for that exact comparison. Vercel wins because its documentation is structured for extraction, something Vercel has discussed publicly as a deliberate strategy (Vercel, 2025). 6sense owns the intent data comparison because its content maps cleanly to the buyer’s decision framework.

This tracks with what the Princeton GEO study found: adding statistics to content improved AI visibility by 41%, the highest single-tactic improvement in the study, while keyword stuffing decreased it by 3% (Princeton KDD, 2024). The improvements were not correlated with domain size. They were correlated with content format.

The 25 unstable queries tell the same story from the other direction. “Best OKR software for B2B companies” has no stable #1. Neither does “best knowledge base software for SaaS” or “best accounts payable automation software.” These are real buyer queries with real revenue behind them, and nobody owns them yet. A company with well-structured comparison and evaluation content could lock in #1 on any of these today.

Giants hold 65% of #1 positions. But 25% of queries have no stable leader at all. The right content structure beats the bigger brand.

Once You Hold #1, You Keep It

The non-determinism story has two layers. On the surface, the data looks volatile: a query surfaces 8.2 unique brands across 10 runs, but only 3.1 appear in every single run. The Jaccard similarity between any two runs averaged 0.72, meaning roughly 28% of the response changes between runs. This is consistent with the broader finding that 40 to 60% of domains cited in AI responses change month-to-month, with drift reaching 70 to 90% over a six-month window (Profound, 2026).

But the top position is far more stable than the rest. For 75 out of 100 queries, the same brand held the #1 recommendation in 70% or more of runs.

This changes the calculus. Position one is a defensible asset. Once earned, it compounds. The brands in positions two through five rotate in and out based on retrieval variance. If your strategy is “get mentioned somewhere in the response,” you are competing for an unstable slot. If your strategy is “own the #1 recommendation for my category query,” the data says that position holds.

Semrush’s own GEO program showed the same compounding pattern: AI share of voice nearly tripled from 13% to 32% in a single month, and non-brand visibility rose from 40% to 50% over the same window (Semrush, October 2025). The top gets stickier. The middle churns.

The #1 recommendation is the same 75% of the time. Positions 2-5 shuffle every run. Own the top or accept the coin flip.

Where Perplexity Pulls Its Citations From, and Why That Is Changing

Across 7,629 total citations in the dataset, the current source distribution skews heavily toward independent publications.

At first glance, this looks like SEO all over again: high-domain-authority publications dominate citations the way they dominate Google rankings. But only 11% of cited domains overlap between ChatGPT and Perplexity (Averi, 2026), which means the publications winning in one engine are not the publications winning in another. Publications still hold outsized citation share inside each engine.

The pattern is already cracking. MarketBetter.ai, a vendor site, publishes a single structured article on outbound sales tools and earns repeated Perplexity citations across multiple runs in our dataset. Not because MarketBetter has Forbes-level domain authority. Because the article is structured for extraction: categorized sections, bullet-pointed feature comparisons, clear brand-to-use-case mappings, and source attributions. Perplexity can parse it cheaply and cite it accurately.

This is the structural shift that separates GEO from SEO. In SEO, domain authority creates a flywheel: high-authority sites rank higher, get more links, and rank higher still. In GEO, citation decisions are driven by extractability and accuracy, not backlink profiles (Princeton KDD, 2024). A well-structured vendor page that answers the query directly can earn a citation alongside, or instead of, a publication that ranks on domain authority alone. Answer engines were built to provide answers, not to recreate the affiliate marketing flywheel.

YouTube appeared in 35.5% of all responses, making it the most common non-blog source, ahead of G2 at 32.3%. Perplexity treats video transcripts as citable content.

Reddit was cited zero times across 1,000 API responses, despite being one of the most commonly cited sources on the consumer Perplexity web product. Reddit has locked down API access and requires paid licensing agreements. OpenAI and Google have signed those deals. Reddit has sued both Perplexity and Anthropic for scraping without licenses (CNBC, 2025). The Sonar API strips Reddit citations. Any GEO tool measuring visibility through the Perplexity API sees a different citation landscape than what end users see on perplexity.ai.

82% of citations come from publications today. But structured vendor content is already breaking through on extractability, not domain authority. The balance is shifting.

Backfire Rates by Article Type

A backfire occurs when Perplexity cites a brand’s own content as a source but recommends competing brands ahead of the content owner in the response. The content gets used. The author does not benefit.

Listicles backfire at nearly 9x the rate of comparison posts. The content format that drove organic traffic in SEO, the “Top 10 Tools” roundup, is the format most likely to help your competitors in GEO.

The industry is still applying SEO content formats to a fundamentally different system. 87% of content marketers plan to increase content marketing budgets in 2026 amid AI search disruption, with one in four now prioritizing LLM models as their primary audience (Clutch and Conductor, 2026), but the default format those budgets fund is still the SEO listicle.

Listicles backfire at 9x the rate of comparison posts. The SEO playbook is actively working against you in GEO.

Two Types of Backfire

Not all backfires are equal. The data reveals two distinct patterns:

Direct competitor backfires (86 instances): The content owner competes in the category they are writing about. Salesforce publishes a listicle about “top sales engagement platforms.” Perplexity cites Salesforce’s article but recommends Outreach, Salesloft, and Apollo ahead. Salesforce wrote the content. Outreach got the recommendation. This is a genuine loss.

Content-play backfires (104 instances): The content owner writes about a category they do not directly compete in. Monday.com publishes “best AI sales tools 2026” to capture traffic across adjacent categories. Perplexity cites Monday.com’s article but recommends Salesforce, HubSpot, and Gong. For Monday.com, getting cited as a source still has brand value. They were never trying to be recommended as an AI sales tool.

The distinction matters for strategy. If you are writing listicles in your own category, the backfire is real and costly. If you are writing in adjacent categories for traffic, getting cited as a source still carries brand value even when the AI recommends someone else.

45% of backfires are direct competitor losses. 55% are content-play brands writing outside their category. Know which one you are.

The Salesforce Example

Salesforce is the most interesting case in the dataset. It appears on both sides of the backfire equation.

As backfire victim: Salesforce’s own content was cited as a source 94 times across the dataset. But when Salesforce published listicles about sales engagement and conversation intelligence, Perplexity recommended Outreach, Salesloft, and Gong ahead of Salesforce in those categories. On “top sales engagement platforms,” Salesforce’s content was cited in every run but Outreach was recommended first in every run.

As backfire beneficiary: Other brands’ listicles helped Salesforce get recommended 41 times. When Monday.com or HubSpot wrote “best CRM” or “best sales tools” listicles, Perplexity used their content as source material but recommended Salesforce based on its dominant third-party citation profile.

The contrast between Salesforce and Gong is notable. Salesforce benefits from backfires 41 times. Gong benefits just once. Yet Gong holds #1 on its own evaluation and comparison queries at 100% consistency. Gong’s content is built for its own category. Salesforce’s content is spread across categories it does not always dominate. Focused content wins the queries that matter.

Salesforce was recommended by other brands’ content 41 times. Gong benefits just once, but owns every query in its category. Focused beats broad.

“Alternatives” Queries Backfire the Most

“Cheaper alternatives to Salesforce” backfired in 7 out of 10 runs. Salesforce’s own content was cited as a source, but Perplexity recommended HubSpot and Microsoft ahead. When you write “[Competitor] alternatives” content, you build a structured menu for the AI to pick from, and it picks based on third-party signals, not your editorial preference.

Evaluation queries showed zero backfires across all 13 queries tested. “Gong AI features review,” “Ramp reviews 2026 enterprise,” “is Ahrefs worth it for enterprise SEO”: when the content focuses on a single product, Perplexity has no structured list of alternatives to reorder. This aligns with the Princeton GEO study’s finding that content with specific statistics, quotations from authoritative sources, and structured claims outperforms generic roundup formats in AI citation (Princeton KDD, 2024).

85% backfire on alternatives posts. 0% on evaluation posts. The safest content in GEO is content about your own product.

Backfire Rates by Vertical

Customer success had the highest backfire rate at 51.2%, driven by Zendesk and Intercom’s dominant third-party presence on G2 and review sites. CRM follows at 26.0%, where Salesforce’s citation profile overpowers most listicle authors. Conductor’s own GEO program delivered a 448% increase in AI citations over several months (Conductor, 2025), and similar review-aggregator dynamics show up in other verticals where one or two third-party platforms dominate the citation surface.

Security and dev tools showed zero backfires, though these had smaller sample sizes (8 queries each). The pattern is consistent with more technical verticals where Perplexity relies on documentation and vendor content rather than listicle roundups.

What This Means for Content Strategy

The pivot is not about writing less. It is about writing differently. Stop building structured lists of competitors for AI engines to reorder. Start writing content that positions your product as the answer to specific buyer questions.

The brands winning in our data (Apollo, Vercel, 6sense, Gong, Highspot) all share the same approach: they own their category query with focused, structured content that leaves no room for the AI to substitute a competitor. That approach is available to any company at any size. The data proves it.

The question is not “what should we write about?” It is “does this content give the AI a reason to recommend someone else?”

How to Choose the Right Content Format for a GEO Query

The choice of article format carries more weight in GEO than it ever did in SEO. The backfire data makes the stakes concrete: listicles backfire 25.7% of the time, comparisons 2.9%, evaluations 0% (Res AI, 1,000-query Perplexity study, 2026). Match the format to the query, not to the legacy SEO playbook.

• If the query is vendor-vs-vendor, write a comparison page. 2.9% backfire rate. The head-to-head framing gives Perplexity less room to substitute a third brand, and comparison queries already skew to cite comparison content.

• If the query is single-product evaluation, write an evaluation page. 0% backfire rate across 13 queries. A page that focuses on one product has no structured menu for the AI to reorder.

• If the query is “[Competitor] alternatives”, do not write an alternatives listicle. 85% backfire rate. You are building the exact menu your competitors get recommended from.

• If the query is broad category discovery and you are the category leader, a listicle is viable but risky at 25.7% backfire. If you are not the leader, write a comparison page or a how-to-choose framework instead.

• If 25 queries in your vertical have no stable #1, prioritize those first. Open positions compound once earned. The same brand holds #1 in 75% of queries at 70%+ consistency once the slot is claimed.

• If your content is already written as a listicle, restructure the page rather than republishing. A comparison table slotted into an existing listicle drops the backfire risk immediately without a full rewrite.

The decision is about query type to format type. Write the format the query pulls, not the format your traffic team already knows how to produce.

Frequently Asked Questions

Why does Perplexity cite a brand’s own content and still recommend a competitor?

Perplexity treats cited content as source material, not as editorial preference. When a listicle surfaces five brands in a structured format, the retrieval layer uses the extracted data to rank candidates and then reorders them against third-party signals (review sites, publication mentions, category authority). The content owner loses control of the order the moment the AI has a menu to pick from. This is why alternatives and listicle formats backfire at 85% and 25.7% while single-product evaluations backfire at 0%.

How big a sample size do I need to detect a stable #1 position?

10 runs per query is the floor for signal; 60 to 100 is the standard for a production monitoring program. The Res AI study ran each of 100 queries 10 times and found Jaccard similarity averaged 0.72 between any two runs, meaning 28% of the response changes per run. A single run cannot tell you whether the #1 position is stable or volatile.

Why did Reddit appear zero times in the dataset when Perplexity is known for citing Reddit?

The Perplexity Sonar API strips Reddit citations that appear in the consumer web product. Reddit has locked down its API and sued Perplexity and Anthropic for unauthorized scraping (CNBC, 2025). Any monitoring tool built on the Sonar API sees a different citation landscape than end users on perplexity.ai. Teams that rely on API monitoring alone are missing the Reddit surface entirely.

Why did customer success have a 51.2% backfire rate when other verticals were near zero?

Zendesk and Intercom hold dominant G2 and third-party review profiles in customer success. When any vendor in the space publishes a listicle, Perplexity cites the listicle as a source but recommends the brand with the strongest third-party signal. The pattern holds wherever one or two brands own the review platforms or review aggregators dominate the surface. Vertical concentration predicts backfire risk.

Why is 25% of queries having no stable #1 the biggest signal in this dataset?

25 queries tested had no brand hold #1 at 70% or higher consistency. These are real B2B queries with revenue behind them, and the position is literally open. A company with well-structured comparison and evaluation content can lock in #1 on any of these in weeks, not quarters. This is a bigger opportunity than fighting for share on a query where a giant is already entrenched at 90% consistency.

How does this dataset compare to ChatGPT citation behavior?

This study is Perplexity-only. Backfire rates may differ on ChatGPT, which pulls from Bing’s index rather than Perplexity’s own crawl. Only 11% of domains are cited by both engines (Averi, 680 million citations, 2026). The listicle-backfire pattern is expected to generalize because it is structural (the AI reorders any structured menu against external signals), but the specific percentages are a Perplexity measurement.

Do content-play backfires actually hurt a brand if they still got cited?

The answer depends on the goal. For a brand writing outside its own category purely for traffic or awareness, being cited as a source has brand value even when the AI recommends a different product. For a brand writing inside its own category, a content-play backfire is a direct loss: their own listicle is training the engine to recommend someone else. Know which bucket the content falls into before publishing.

What does “focused content” look like in practice for a mid-size brand?

Gong is the clearest example in the dataset. Gong holds #1 on its own evaluation and comparison queries at 100% consistency, and benefits from other brands’ backfires only once. Its content is built exclusively around its own category (revenue intelligence, conversation intelligence), with comparison pages and evaluation pages that give Perplexity no open menu to reorder. Salesforce, by contrast, spreads content across categories it does not always dominate and benefits from backfires 41 times while losing its own listicle queries.

How fresh does cited content need to be to compete?

Recent, but not weekly. AI-cited content is 25.7% fresher than traditionally ranked content on average (Ahrefs, 2025). A quarterly refresh of the core stats, comparison tables, and pricing data is enough to stay on the fresh side of the distribution. Updating less frequently means an article with 2024 data loses to one with 2026 data even when the 2024 article is better written.

Is a listicle with 100 items safer than one with 10?

No. The backfire mechanism is the structured menu, not the menu length. A 50-brand listicle gives Perplexity 50 candidates to rank against third-party signals instead of 10. The fix is not more entries; the fix is moving from listicle format to comparison, evaluation, or how-to-choose structure for the queries that matter most.

Methodology

Query selection: 100 B2B software queries spanning discovery (“best X tools”), comparison (“X vs Y”), evaluation (“is X worth it”), and alternatives (“X alternatives”). 10 verticals: AI, CRM, customer success, data, dev tools, finance, HR, marketing, project management, security. Sample sizes ranged from 8 to 15 unique queries per vertical.

Execution: Each query was run 10 times through Perplexity’s Sonar API (temperature 0.7) to account for non-deterministic responses. Total dataset: 1,000 results with 7,629 total citations analyzed across 739 unique domains.

Backfire metric: A backfire occurs when a brand’s own content is cited as a source in Perplexity’s response, but the response recommends competing brands ahead of the content owner. Citation ownership was determined by matching cited URLs to brand domains. We distinguish between direct competitor backfires (owner competes in the category) and content-play backfires (owner writes in adjacent categories for traffic).

Reddit and API limitations: The Perplexity Sonar API does not surface Reddit citations, even though the consumer Perplexity web product cites Reddit frequently. Reddit requires paid licensing agreements for API data access and has sued Perplexity and Anthropic for unauthorized scraping (CNBC, 2025). Any GEO measurement through the Sonar API reflects a different citation landscape than end users see on perplexity.ai. This limitation applies to any tool or platform using the Perplexity API for monitoring.

Other limitations: This study covers Perplexity only. Backfire rates may differ on ChatGPT, Claude, and Gemini. Vertical sample sizes are uneven (8-15 queries each). Temperature 0.7 introduces controlled variance; production Perplexity responses may vary.

Sources cited: Princeton KDD 2024 (GEO optimization strategies), Profound 2026 (AI citation drift), Averi 2026 (ChatGPT vs Perplexity citation overlap), Ahrefs 2025 (citation freshness analysis), Semrush October 2025 (AI share of voice compounding), Conductor 2025 (AI citation growth), Clutch and Conductor 2026 (content marketing budget survey), CNBC 2025 (Reddit v. Perplexity lawsuit), Vercel 2025 (AI search optimization approach).

Res AI is the autonomous GEO engine for your CMS. It connects to WordPress, Webflow, Framer, or Contentful with a simple login, monitors your AI citations daily, and deploys content where you are not being cited. The architecture is built for comparison and evaluation content, the formats this data shows carry the lowest backfire risk and the highest citation control.

See how it works →