Category Gravity
The reference layer.
We publish the reference layer a category checks itself against.
Research, not advice. We establish what is true, show the work, and hand you the tools to apply it. Every claim carries its primary source and its date. First-party disclosures rank first. A number without a source is treated as a marketing asset, not a fact.
Verdict grammar
Three words, every report.
So you learn to read us fast. The same verdict means the same thing whether it lands on a reach stat or a penalty rumor.
Verified
A primary source supports the claim as stated. Safe to restate behind your own name.
Contested
A real effect, but the magnitude is wrong, unsourced, or it flips by context. Use the finding, drop the number.
Myth
It traces to no auditable source, or a primary source contradicts it. Repeated confidently, backed by nothing.
How to read a Category Gravity report
Primary versus inferred.
Every report splits what a primary source actually discloses or measures from what we infer on top of it. The first kind you can restate behind your own name and defend in a room. The second is a hypothesis, and it is marked as one. That line is the whole point of the imprint. It is what lets you apply the work to your own situation without misapplying it. When a report says a mechanism is confirmed but a figure attached to it is not, both facts survive on the page. We do not launder the shaky number into the solid one to make a cleaner story.
The canon
Numbered, dated, newest first.
Every report is sourced, dated, and re-checked. Newest at the top.
Research 006
August 2026 · As-of dated, re-checked
The AI Trust Gap: Who AI Recommends When a B2B Buyer Asks
Original measurement, not a synthesis. We put 15 questions a growth-stage B2B founder asks before buying to five AI engines, three times each, and logged every cited source. 632 citations later, the two engines most buyers use, ChatGPT and Claude, cite McKinsey, a16z, Y Combinator, and HBR, not specialists. The barrier to being recommended by AI is entity trust, not answer quality.
Research
The report
The sourced report. The most-cited domains on the strict engines, the engine split, and why AI defaults to incumbents, from a fixed, reproducible probe.
Read the full report
Synthesis
The playbook
The entity-authority playbook: how a growth-stage company earns citations from the engines that default to household names.
Planned.
Application
The audit
The AI citation audit, run on your own domain and buyer questions.
Planned.
Serialization forthcoming in the
Category Gravity newsletter as "I Checked the AI Citations."
Research 005
August 2026 · As-of dated, re-checked
Can Buyers Verify Your Market-Leading Claim?
A founder-facing review of ten beliefs about how B2B market claims get checked. Public evidence can show whether a claim is findable, current, consistent, supported, or contradicted across search, AI answers, customer proof, reviews, competitor material, and approval documents. It cannot reveal a buyer's private motive or predict a committee's decision.
Research
The report
Ten claims ruled verified, contested, or myth against B2B buyer studies, professional research standards, and decision-quality guidance.
Read the full report
Synthesis
The playbook
A working sequence for testing one claim across one named buying group before the company funds and repeats it.
Read the playbook
Application
Claim and Date
Bring one claim and one decision date. Get a portable pre-read of the evidence questions that still need answers.
Planned.
Serialization forthcoming in the
Category Gravity newsletter as "I Checked the Market-Leading Claim."
Research 004
August 2026 · As-of dated, re-checked
Pricing Myths: Charm Pricing, Tiers, and the Numbers Founders Repeat
A founder-facing audit of the pricing tactics everyone repeats. Charm pricing, the three-tier decoy, the "1% price beats volume" statistic, value-based pricing, and the SaaS discount conventions, each checked against the peer-reviewed experiments they trace to. A few have real field studies behind them. The catch is always in the condition the popular version drops.
Research
The report
The sourced report. Eleven claims ruled verified, contested, or myth against the original pricing and decision-science literature, plus the buried assumptions the viral versions leave out.
Read the full report
Synthesis
The playbook
The report's findings turned into an operating system for founders setting their own prices.
Read the playbook
Application
The tool
An interactive check for the pricing claim you are about to build a page around.
Planned.
Serialization forthcoming in the
Category Gravity newsletter as "I Checked the Pricing Numbers."
Research 003
August 2026 · As-of dated, re-checked
Is SEO Dead? Search, AI Overviews, and Zero-Click in 2025-2026
A founder-facing audit of the "SEO is dead" panic. Zero-click search, AI Overviews, "Google is losing to ChatGPT," and the decade-old ranking-factor stats, each checked against Google's own disclosures, Alphabet's filings, and neutral clickstream panels. Google is bigger than ever and the click economics still changed. The honest read is that the return moved from ranking-for-traffic to being the answer.
Research
The report
The sourced report. Eleven claims ruled verified, contested, or myth, a primary-versus-inferred split, and a recorded verification pass, including the neutral Pew panel and Alphabet's own filings.
Read the full report
Synthesis
The playbook
The report's findings turned into an operating system for founders deciding where search still pays.
Read the playbook
Application
The tool
An interactive check for the SEO stat you are about to act on.
Planned.
Serialization forthcoming in the
Category Gravity newsletter as "I Checked the SEO Numbers."
Research 002
August 2026 · As-of dated, re-checked
Cold Email Benchmarks and Myths in 2025-2026
A founder-facing audit of cold email benchmarks. Reply rates, personalization multipliers, send-time folklore, deliverability rules, and the legal panic, each checked against transparent large-sample datasets and primary platform and regulator documents. Most viral numbers trace to one vendor, a 2013 study about a different kind of email, or no source at all. The finding underneath: reply rate is denominator theater, and the honest levers are boring.
Research
The report
The sourced report. Ten claims ruled verified, contested, or myth, a primary-versus-inferred split, and a recorded verification pass on every number the first draft could not source.
Read the full report
Synthesis
The playbook
The report's findings turned into an operating system for founders running outbound.
Read the playbook
Application
The tool
An interactive check for the cold email stat you are about to act on.
Planned.
Serialization forthcoming in the
Category Gravity newsletter as "I Checked the Cold Email Numbers."
Research 001
August 2026 · As-of dated, re-checked
LinkedIn Organic Reach in 2025-2026
A founder-facing audit of the LinkedIn organic reach panic. The repeated stats about throttling, silent penalties, golden hours, and dwell-time cliffs, each checked against LinkedIn's own engineering disclosures and transparent, large-sample third-party datasets. Most of the scariest numbers trace to reports that never expose their method.
Research
The report
The sourced report. Claim-by-claim verification, the primary sources with a note on why each is credible, and the split between what is measured and what is inferred.
Read the full report
Its verified findings are also live in the application layer.
Synthesis
The playbook
The Founder's LinkedIn Playbook. The report's findings turned into an operating system for the founder who wants to act, not just know what is true.
Read the playbook
Application
Reach Reality
Check any LinkedIn claim against its source and watch the verdict resolve. Backed by a dated evidence library of 21 repeated claims.
Open the checker
Read the claims index
Serialized publicly, one claim at a time, in the
Category Gravity newsletter. The full report is the depth layer behind each issue.
AI Assurance research
How long does it take a client to say yes?
The AI Permission Cycle Time method defines how Petrichor will measure the time, labor, repeated diligence, evidence reuse, decision result, and control preservation behind one bounded AI-use decision. No results have been collected.
Inspect the proposed research method