TL;DR

PostHog's homepage now promises AI agents that "automatically generate pull requests," but the site scores only 69/100 and the pivot creates three verified revenue leaks. Despite a 4.5/5 G2 rating from 1,000+ reviews, that trust signal is absent from the homepage and pricing page. The pricing page lists 17 separately metered products, making a total bill impossible to estimate without leaving the page—a conversion drag independently corroborated by third-party cost breakdowns.

The bottom line: for cold, non-technical visitors, these friction points create a modeled 10-15% drag on signup completion, and the article's verdict is that PostHog should surface its existing social proof and ground the AI pitch with a clearer "what this is" before the "what it does."

PostHog Website Review: 3 Revenue Leaks Behind the "Self-Driving" Pivot

We published an earlier review of posthog.com that scored the site on messaging, conversion, and trust. When we re-audited it this week, we found something more interesting than confirmation: the site had changed underneath the old critique. PostHog's homepage no longer says what our previous review said it said, its case-studies page is not what we described, and at least one "trust gap" we called out doesn't exist. So this is a full rewrite, built only from what we could verify live on posthog.com, posthog.com/pricing, and posthog.com/customers this session, plus third-party corroboration. Where the old claims turned out to be wrong, we say so explicitly below.

Overall Score: 69/100

Executive Summary

Direct answer: PostHog is now selling a different story than it was. The homepage headline is "Shift your product into self-driving mode," with sub-copy promising PostHog "automatically diagnoses problems, fixes bugs, and generates pull requests — all without you having to prompt it" — a pivot toward an AI-agent narrative built on top of its existing analytics/flags/replay stack (posthog.com). That's a bold, specific claim, and it's genuinely different from generic analytics-tool copy. It also creates three verifiable friction points for self-serve conversion:

  • Messaging leak — the boldest, least-established claim (AI agents writing your pull requests) is now the first thing a cold visitor reads, ahead of any grounding in what PostHog fundamentally is (a product analytics, flags, and session-replay platform). For a PM or growth lead evaluating tools, "does this even do what I need" isn't answered until well after the AI-agent pitch.
  • Conversion leak — the pricing page is transparent about per-product usage-based pricing, but sizing an all-in monthly bill across multiple products (analytics + replay + flags + surveys, each metered separately) is genuinely hard to do without leaving the page, and this is independently corroborated by multiple third-party cost breakdowns, not just our read of the page.
  • Trust leak — PostHog has strong, verifiable trust signals (a 4.5/5 G2 rating across 1,000+ reviews, recognizable customer logos including Y Combinator and Supabase) that simply are not surfaced on the homepage or pricing page itself. The proof exists; it's just off-site.

We model these as a combined 10-15% modeled drag on self-serve signup completion for cold, non-technical visitors specifically — this is nqzai's own estimate based on the friction points below, not a measured PostHog statistic, and we don't have access to PostHog's actual funnel data to confirm it.

One important correction to our earlier review: it described the homepage headline as "Build better products," described the site as lacking a customer-logo wall, and put case studies at "6 months old." None of that matches what's on the site now — either the site changed, or those claims were wrong when written. We can't tell which, and we're not going to guess. We're only standing behind what we verified this session.

Messaging Score: 62/100

What we verified: The homepage hero is "Shift your product into self-driving mode," followed by "PostHog already knows your customers, which features they use, and the issues they have," then the automation pitch about diagnosing problems and generating pull requests without prompting. Below that, a section header reads "Built-in tools for your agents," introducing a list of 17 products (Product Analytics, Web Analytics, Session Replay, Feature Flags, Experiments, Surveys, Error Tracking, Managed Warehouse, CDP, Workflows, Logs, AI Observability, Endpoints, Inbox, Traces, Heatmaps, Replay Vision). The primary CTA is "Get started – free," alongside "Demo" and "Talk to a human" (posthog.com).

Why this costs conversions: Leading with "your product ships its own pull requests" is a strong differentiator for an engineering audience already fluent in PostHog's category, but it's a hard landing for anyone evaluating PostHog as "just" an analytics or feature-flag tool — which is still how most of its 17 listed products are used per G2's own reviewer base (81% small-business, using it primarily as web/product analytics and flags) (G2 PostHog Reviews). The self-driving framing is also PostHog's own current top-line positioning, documented in company material like its self-driving FAQ (posthog.com/docs/self-driving/faq) and its GitHub description, which now reads "the leading platform for building self-driving products" rather than a product-analytics description. That's a real, deliberate repositioning, not something we're inferring.

What we could not verify: We could not render the page's full interactive behavior (animations, scroll-triggered copy, any A/B-tested variants) via automated fetch, so this reflects the static content served, not necessarily every visitor's exact experience.

Conversion Score: 68/100

What we verified on-page: The pricing page headline is "PostHog pricing – Transparent, usage-based, generous free tier," opening with "PostHog is designed to grow with you. Our 10+ products (and counting) will take you from idea to product-market fit to IPO and beyond." The free tier requires no credit card and includes meaningful allowances (1M analytics events, 5K session recordings, 1M feature-flag requests, 1.5K survey responses, 1M data-warehouse rows, one project, one year of retention). Adding a card unlocks six projects, seven-year retention, and email support. Pricing is tiered by volume per product — e.g., product-analytics events start around $0.00005 each and drop with volume — and each product (replay, flags, surveys, warehouse) is metered and priced independently rather than bundled (posthog.com/pricing). The CTA is "Get started - free," linking to signup.

What we could not verify: The pricing page references a "Pricing calculator" section, but our automated fetch could not render its interactive slider behavior (it's JavaScript-driven), so we can't confirm exactly what a user sees when adjusting inputs, or whether it surfaces one combined estimated monthly total across products. Similarly, we could not fully render the signup form itself (app.posthog.com/signup redirects to a JS single-page app), so we're not able to independently verify field count or steps — we're flagging this rather than guessing at specifics, unlike claims of "minimal fields (email + password)" in the prior version of this review, which we cannot confirm and are dropping.

Third-party corroboration: Independent cost breakdowns describe exactly the friction pattern you'd expect from per-product metering: "the same tool shows up in one Reddit thread as 'basically free' and in the next as a '$60k/month' bill," largely because "PostHog prices each product separately rather than bundling everything into one broad plan," and because identified (logged-in) events can cost up to 4x anonymous ones — a distinction one buyer-side analysis calls "the catch nobody flags" (BudgetForge, PostHog Pricing 2026). PostHog has responded to cost pressure before: it cut session-replay pricing substantially (25K recordings dropped from $430/month to $85/month), attributing the change to pipeline optimization rather than to complaints specifically (PostHog: "We've decided to make less money"). That the company felt the cut was worth publicizing at all suggests replay cost was a live objection in sales conversations, even if the post itself doesn't say so.

Correction to the prior review: The earlier version claimed an "Enterprise tab" on pricing creates confusion for self-serve users. We found no Enterprise tab or contact-sales flow anywhere on the current pricing page — it's Free and Pay-as-you-go only. If that tab existed before, PostHog appears to have simplified it away; either way, the claim doesn't hold today.

Trust Score: 76/100

What we verified: posthog.com/customers is a dedicated, populated case-studies page, not a thin afterthought — it features named, recognizable companies including Y Combinator, Supabase, ElevenLabs, ResearchGate, and Exa, each tagged with which PostHog products they use (Experiments, Product Analytics, Feature Flags, etc.) (posthog.com/customers). The homepage itself also carries a social-proof line — "Yes they actually use us, no it's not just some random engineer who tried us out 2+ years ago" — next to a customer logo strip. Independently, PostHog holds a 4.5/5 rating on G2 across roughly 1,000-1,050 reviews, with specific sub-scores like 8.9/10 for "meets requirements" and 92% for "ease of doing business with" (G2 PostHog Reviews).

The actual gap: none of that G2 data — the rating, the review count, a badge — appears on the homepage or pricing page itself. A visitor comparing PostHog to a competitor has to leave the site and go find the G2 page themselves to discover the strong third-party validation exists. We also found no visible Capterra or TrustRadius badges on-site. That's a real, verifiable gap, distinct from the old review's now-incorrect claim that there's "no customer logo wall" and "only single quotes from unnamed sources" — there clearly is a logo wall and a real case-studies page; it's the external rating proof that's missing on-site, not the customer proof.

Where PostHog is candid, to its credit: the self-driving FAQ proactively addresses the obvious trust objection to letting an AI agent touch your codebase — "Every change is a pull request a human reviews... nothing merges itself," and on the business model, "You pay per PR, and we refund PRs that miss the bar. A business model that charges for slop dies quickly" (posthog.com/docs/self-driving/faq). Addressing the objection before the visitor has to ask it is a trust-building move, not a trust gap — we're noting it as a genuine strength, not an assumption.

Independent friction, for balance: third-party review roundups consistently name a steep learning curve as PostHog's most common complaint — "the #1 complaint across G2, Product Hunt, and Reddit," per one 2026 review aggregation, with reviewers noting the tool "can feel overwhelming at first" for less technical users (CubeAPM PostHog Review 2026). That's a real, sourced trust/usability signal working against self-serve activation, separate from anything about the website itself.

Recommendations

  1. Re-sequence the homepage narrative. Keep the self-driving pitch — it's differentiated — but put one line ahead of it that plainly states the category ("product analytics, feature flags, session replay, and AI agents that act on all three") so non-technical evaluators aren't lost before they know what they're looking at.
  2. Surface a single blended cost estimate. Since per-product metering is the most consistently cited pain point across independent sources, add a one-line "estimated total monthly cost across the products you selected" summary above the individual product tables on the pricing page — not just per-product breakdowns.
  3. Call out the identified-vs-anonymous event cost distinction explicitly, ideally right in the calculator, since it's the specific "hidden cost" third parties flag most.
  4. Import the G2 rating onto the homepage or pricing page. A verified 4.5/5 across 1,000+ reviews is being wasted by living only on a page most visitors never navigate to.
  5. Re-verify before quoting. For any future audit of this site (ours or anyone else's), fetch the live page first — this review exists because the previous one didn't, and several of its specific claims (headline copy, missing logo wall, an Enterprise tab) turned out not to match reality when checked.

Sources