# Marketing takeaways Speaker notes for the slides. Repository retrospective and PostHog history through 2026-09-27. Source evidence is not a claim of business lift. ## 1. The product built the story. The data sharpens the next move. The codebase reveals a connected marketing system across Gateway, DevPass, Lounge and Airside. This updated presentation adds seventeen months of PostHog history, from May 2025 to today, plus search and audience snapshots. Financial data and personal records are excluded. The findings sharpen the next decisions. They do not establish causal marketing lift. Source: Repository 01da451e5c75 · analytics captured 2026-09-27 ## 2. Four products. Four reasons to arrive. Gateway serves engineering teams. DevPass serves developers using coding agents. Lounge serves people using AI directly. Airside serves inference providers. Shared navigation connects the family. Each audience still needs its own evidence and route to first success. The concurrent homepage refresh is not treated as a verified production release. Source: Product surfaces · shared switcher · homepage refresh remains local work ## 3. A substantial archive. Not a growth metric. The snapshot contains 245 editorial files and 143 product documentation pages. Announcements, comparisons, setup guides and migration paths create a substantial owned foundation. These counts establish work completed. They are not visitors, customers or proof that a channel performs. Source: Frozen repository inventory · evidence.json ## 4. A release becomes useful when the path is complete. The reusable pattern is release, explanation, proof, setup and qualified activation. The repository also contains launch records, product tours, newsletter capture, lifecycle nudges and referrals. Implementation is only one evidence level. Publication, delivery and successful use still require their own checks. Source: Blog + changelog + docs · product tours · newsletter and referral implementations ## 5. Three sources. Three different clocks. PostHog now covers the full history, from the first tracked day in May 2025 to today, counted as people per month. It also supplies one detailed week-on-week comparison. Search Console uses its native seven day view. Admin supplies all time audience totals. Keep those scopes separate. A person counted in several months is counted in each, and all time account totals do not belong in a monthly funnel. Source: analytics.json · source filters, capture times and query definitions ## 6. Two jumps. Only the second one held. In June 2025, 1,522 people visited. In the first twenty seven days of September 2026, 37,011 did. The July 2025 launch reached 9,441 people, then fell to 3,148 two months later. The second jump came in July 2026, with DevPass and the coding plans article. That one held and kept growing. Source: analytics.json · PostHog MCP · 2025-05-01 to 2026-09-27 UTC ## 7. Google compounds. Social does not. People arriving from Google grew every quarter, from 2,985 in Q3 2025 to 40,071 in Q3 2026. Social and community referrals peaked in the launch quarter at 5,812 and stayed near two thousand four hundred a quarter for a year. Referrers describe pageviews, not first touch. Direct also includes bookmarks and stripped referrers. Source: analytics.json · PostHog MCP · 2025-05-01 to 2026-09-27 UTC ## 8. Catalogue pages and tools grew. Comparisons stalled. Model pages grew from 1,909 to 9,276 people a month. Tools like the timeline, rankings and calculators went from 22 to 5,396. Compare and migration pages moved from 709 to 611. They have been flat since February. Pages generated from the live catalogue keep earning search traffic. One-off comparisons need a refresh before more are written. Source: analytics.json · PostHog MCP · 2025-05-01 to 2026-09-27 UTC ## 9. Small, but the fastest-rising source. AI assistant referrals rose from 10 people in June 2025 to 450 in September 2026. Gemini alone went from 10 to 180 in six months. ChatGPT also tags its links, adding 2,237 people the referrer count misses. Merge both signals into one channel. Source: analytics.json · PostHog MCP · 2025-05-01 to 2026-09-27 UTC ## 10. Newsletters worked once each. Newsletter mentions such as TLDR Founders, with 2,092 people, brought real visitors, once. None of them kept sending people after the issue went out. Plan every launch or mention around a page that keeps earning search traffic afterwards. Source: analytics.json · PostHog MCP · 2025-05-01 to 2026-09-27 UTC ## 11. Fix the signals before the funnel. PostHog's sign-up event covers about 16 percent of the sign-ups in the admin panel. 52 percent of answers to "where did you hear about us" are "Other", because the options leave out the biggest sources. Click tracking only started in March 2026. Fix these signals before measuring any conversion rate. Source: analytics.json · PostHog MCP · 2025-05-01 to 2026-09-27 UTC ## 12. The latest week: attention held, activity dipped. Production pageviews fell 22.5 percent, while tracked visitor identifiers fell 3.1 percent. CTA click events fell 35.1 percent and recorded signup identifiers fell 21.3 percent. These are observed adjacent week movements, with different event populations. They are not an experiment or a causal verdict. Source: analytics.json · PostHog · 20 Sept–26 Sept vs 13 Sept–19 Sept 2026 UTC ## 13. Investigate DevPass before adding more assets. DevPass pageviews moved from 39,509 to 25,018, accounting for 80.8 percent of the measured decline. Gateway visitor identifiers were nearly flat. Check the acquisition mix, returning activity, instrumentation and first success cohorts before blaming a campaign or a release. Source: analytics.json · PostHog · 20 Sept–26 Sept vs 13 Sept–19 Sept 2026 UTC ## 14. Demand is visible. Search lift is not established. The search summary shows real demand, and brand queries lead the visible rows. Additional discovery includes provider and alternative related terms. Keep the headline counts rounded, as the console displays them. A top query sample does not establish total brand share. A consistent comparison window is still needed before claiming SEO improvement. Source: analytics.json · Search Console rendered summary and visible query rows ## 15. Account totals are not weekly acquisition. The all time Audience panel shows 23,308 signups, 20,225 verified users and 30,542 organizations. Verification is not first use activation. Organizations are not unique people. These totals add context, but must remain separate from the weekly event cohort. Financial panels were excluded. Source: analytics.json · Admin Audience · observed 2026-09-27 ## 16. The same evidence must reach every artifact. A report refresh does not update an already rendered video. Keep the report, deck, spoken script, narration, captions and final media on the same approved evidence revision. Regenerate the affected outputs and verify the actual files. The same privacy limits apply to text, imagery and supporting evidence. Source: Evidence revision 614f65f170e2 · presentation-manifest.json ## 17. Feed search. Fix signals. Then experiment. Keep investing in pages built from the live catalogue, and refresh them whenever models ship. Rework the stalled comparison pages around current search intent. Fix the sign-up event and the source survey, and merge AI assistant signals into one channel. Then test one meaningful change with comparable windows and quality guardrails. Keep the failures in the learning record. Source: analytics.json · full-period decisions and remaining uncertainty ## 18. Ground the promise. Keep the whole story current. The upgraded marketing skill now covers authorized analytics, financial exclusions, time windows, identity units and synchronized deliverables. The public pull request contains generalized methods, not this account's metrics or private dashboard links. Market what the product can prove, measure what changed, and keep every requested artifact current. Source: SoulStack PR #2 · marketing skill · no account metrics in the public contribution