TL;DR — the core findings

The Query and What Came Back

The exact query was: "best free SaaS directories 2026." Perplexity returned ten cited sources. I went through each one looking for a pattern.

Every single source had the same structure. Not the same design — the same information architecture. Each cited page led with a scannable list or table where specific claims appeared in labeled, extractable form:

One competitor appeared three times across the cited sources — not because three separate sites independently chose to feature it, but because that competitor had published a dedicated standalone comparison page with a proper HTML table. Perplexity had cited its own comparison article, plus two other sites that had quoted that same table. One piece of structured content generated three citations.

What the cited pages had in common

Every source Perplexity cited for this query was a page where the core facts appeared in a labeled, structured format — not buried in descriptive prose. None of them were homepages. All of them were either comparison articles or dedicated "best of" roundups with explicit methodology sections. This is the content format AI engines are built to extract facts from.

What Structured vs. Narrative Content Looks Like to an AI Engine

The gap between cited and non-cited content isn't about length, depth, or even accuracy. It's about extraction cost — how much parsing effort it takes for an AI engine to pull a specific, quotable fact from the page.

Dimension Narrative-Only Content Structured Content
How facts appear Buried in descriptive paragraphs ("we have a strong domain rating and fast approvals") In labeled rows or cells ("Domain Rating: 86 · Approval: instant")
Metric format Qualitative descriptors: "high," "fast," "trusted," "comprehensive" Quantified with units: "DR 86," "live in <60 seconds," "founded 2023"
Methodology Absent or implied ("we curate the best tools") Explicit section explaining how each metric was verified
AI extraction effort High — engine must parse context, infer meaning, resolve ambiguity Low — engine reads a labeled value directly, minimal parsing required
Citation likelihood Low — AI skips to the next result with cleaner structure High — engine can quote a specific cell or row as a standalone fact
Competitor multiplier One page, one potential citation One table → quoted by multiple sites → multiple citations from one source

What "AEO Friction" Means in Directory Content

Definition — AEO / GEO Friction in Directory Content

AEO friction (in the context of answer engine optimization and generative engine optimization) is the gap between what a page says and what an AI engine can confidently extract as a standalone, citable fact. A high-friction directory page uses vague descriptors ("comprehensive," "fast," "trusted"), embeds key metrics inside paragraphs, and has no methodology section. A low-friction page states claims explicitly with units ("approval: same day," "domain rating: 86," "dofollow: yes"), groups them in a structured table or labeled list, and includes a clearly labeled methodology section explaining how each figure was verified. AI engines resolve friction by skipping the high-friction source and retrieving the next most structured result — regardless of which site is technically more authoritative on the topic.

This is the specific mechanism that explains why ToolIndex wasn't cited even though it had real, accurate information on the same topic. The facts existed. The extraction cost was too high. The competing pages had lower extraction cost, so Perplexity cited those instead.

The Fix That Changed the Citability

The change was targeted: we rebuilt the comparison content with three specific additions that the cited pages all had.

1. A real HTML comparison table

Not an image of a table, not a description of a table — an actual <table> element with labeled rows for domain rating, approval time, dofollow status, submission type, and listing permanence. Each cell contains a specific, verifiable value. The methodology section below the table explains exactly how each metric was measured.

2. Concrete units on every metric

Every qualitative claim got replaced with a quantified equivalent. "Fast approval" became "listings go live in under 60 seconds." "High-authority backlink" became "DR 86 dofollow backlink." "Free" became "free, no account required, no approval queue." Qualitative claims aren't citable. Quantified claims are.

3. An explicit methodology section

A short, clearly labeled section explaining how the domain rating was verified (Ahrefs, snapshot date), how approval times were measured (personal submission tests, documented), and what "free" means in each case (free forever vs. free trial). Perplexity cites methodology sections disproportionately — they're the part of a comparison page an AI engine can treat as a reliability signal, not just a data source.

The structural change, not the content change, is what matters for AI citability. The facts were already accurate before the rebuild. The AI engines weren't finding them because the facts were written for human readers, not for machine extraction. Structured content is content that serves both audiences at once.

The Second Finding: Named-Entity Collisions Are Not Hallucinations

While running these Perplexity tests, I also queried our own proprietary terminology — the Decision Friction Model, the analytical framework behind how ToolIndex and the Strategic Flow audit tool score email and landing page effectiveness.

In one query, Perplexity cited it correctly: the right context, the right application domain (email copy and SaaS landing pages), with a citation pointing to our own content. Good.

In a second query, phrased slightly differently, the answer merged our Decision Friction Model with a completely unrelated "Decision Friction Model" used by a leadership consulting firm. The AI produced a response that combined both — accurate facts about each entity, but attributed to a single entity that doesn't actually exist. This is not a hallucination. The AI retrieved real content from two real sources and merged them because the entity name was identical.

This is a named-entity collision. It's more common than most SaaS founders realize, and it's particularly dangerous because:

How named-entity collisions happen

An AI engine encounters two pages that use the same term — "Decision Friction Model" in this case — in different contexts. Without a disambiguation signal (an explicit statement that your use of the term is distinct from another entity's use), the engine may merge both sources into a single entity in its synthesized answer. The same dynamic applies to generic product names, common acronyms, and any proprietary term that shares its name with an unrelated concept in another industry.

The disambiguation fix

The fix for a named-entity collision is not a legal notice or a request to the AI company. It's content: a clearly labeled disambiguation block on your own site and on your directory listings that states what your term means, what it does not mean, and — where possible — links to third-party content that confirms the attribution.

For the Decision Friction Model: the fix was a short paragraph on our methodology page stating explicitly that the Decision Friction Model as used by ToolIndex and Strategic Flow refers specifically to email and landing page copy analysis, is distinct from any leadership or organizational decision-making framework, and was developed independently for the SaaS content audit use case. After that content was indexed, the next Perplexity test returned the correct attribution in both query phrasings.

The General Lesson for SaaS Founders

AI search engines don't browse your product like a human does. They retrieve structured content, resolve entities by name, and synthesize answers from the sources with the lowest extraction cost. The implications are practical:

Where to Start

Before building a comparison table or writing disambiguation content, do the audit that surfaces whether you have a problem. It takes fifteen minutes:

  1. Open Perplexity. Search the category query your ideal customer would use — not your product name, the problem they're trying to solve.
  2. Read the cited sources. Are you there? If not, read what the cited sources have structurally that yours doesn't.
  3. Search your exact product name. Read the description Perplexity returns. Is it accurate? Is it yours, or partially someone else's?
  4. Repeat in ChatGPT (web search enabled) and in Google (look for the AI Overview at the top of results).
  5. If anything looks wrong or missing, you now know exactly what to fix — and it's a content problem, not a traffic problem.

Most founders have never run this test on their own product. The fifteen minutes it takes to check what an AI model currently says about you will tell you more about your AI search visibility than any analytics dashboard — because it shows you the actual output, not a proxy metric.