Dashboard mockup: Search (SEO) and AI answers (AEO)

1. Where it sits

1. The existing admin dashboard stays exactly as it is. Nothing in it is removed or changed.
2. Below it, one new section with two tabs: Search and AI answers.
3. Headline tiles and tables use the last 28 complete days all sources share (in this mockup: 20 Sep to 17 Oct 2026). Weekly charts show the last seven Monday-weeks (31 Aug to 17 Oct). Every block shows its source's "data through" date; Search Console runs about two days behind.
4. A link to the latest dated bi-weekly report sits at the top of the section (Phase 2).
5. A simple change log (date, page, change, owner) shown as markers on the charts, in Phase 1 if it fits, otherwise Phase 2.

2. Tab one: Search (SEO)

Headline Live 20 Sep to 17 Oct 2026

Authority Dated snapshot, 20 Oct weekly from Phase 2

Ahrefs' traffic estimate and keyword count are left out: Search Console gives the real clicks, and the keyword map tracks our own keywords.

3. Tab two: AI answers (AEO)

How often ChatGPT, Perplexity, Gemini, Copilot and Google's AI answers name Hypertune for our 25 tracked questions, whether what they say is right, and how many people they send us.

Headline

4. What is taken off or moved

From the current admin dashboard

Nothing. No existing block is removed or changed; the new section only adds.

From the 1 October SEO spec

5. Definitions for the build

1. Dates: labels show inclusive start and end dates (20 Sep to 17 Oct 2026 is 28 days). Timestamp filters use UTC, inclusive start, next-day exclusive end (from 2026-09-20T00:00Z to before 2026-10-18T00:00Z). Search Console filters use stat_date between 2026-09-20 and 2026-10-17. Hide comparisons where the earlier period has no data (the survey answers start 29 Sep).
2. Search Console: totals from grain = site only; branded and non-branded from grain = query only, using the brand rule h[iyu]p[eo]?r?\s?t[uo]+n|hyoertune|compready, case-insensitive. Undisclosed = site minus disclosed query clicks and impressions. Never get non-branded by subtracting branded from site totals.
3. Brand trust searches: branded queries that also match legit|review|work|worth|safe|free|crack|scam. Click rate = their clicks ÷ their impressions; target 10%.
4. Rates: click rate = sum of clicks ÷ sum of impressions; position = impression-weighted average. Never average daily rates.
5. Sessions: the earliest entry page view per visitor and session, on hostname hypertune.gg or www.hypertune.gg only (exact match), excluding internal and test traffic. Materialise daily session entries or index (visitor_id, session_id, created_at) where is_entry, so the query stays fast.
6. AI assistants: exact hostname match against a maintained list: chatgpt.com, chat.openai.com, perplexity.ai, gemini.google.com, bard.google.com, copilot.microsoft.com, claude.ai, grok.com, plus utm_source = chatgpt.com. Paid traffic means utm_medium cpc, ppc, paid, paid_social, display, sponsored or affiliate, or a gclid, fbclid or twclid; check AI assistants before any other UTM rule. Google's own AI answers can't be separated from normal Google referrals.
7. Survey: filter by source_id (5 = Online search, 6 = AI Chat), earliest answer per clerk_id, labelled as answers. A Stripe ID or is_trial = false does not prove payment.
8. New payers: first successful, positive payment per Stripe customer, dated in the period; trial conversions count when they first pay. First observed source = the customer's earliest tracked landing before signup. Always show the match rate. Check how many visitors link to a Clerk or Stripe customer before building this.
9. Page groups: trust, game, comparison, guide and homepage, by URL path rules.
10. Keyword map: keep cluster, priority and horizon from the sheet. Totals and weighted position from grain = query over the period; the ranking page is the single page with the most impressions for that query. "Moved into the top five" = weighted position 4 to 20 in the previous period and 5 or better now, seen in both.
11. Indexing: latest inspection per URL, synced daily from the sitemap, with content pages separate from legal and utility pages. Show the oldest inspection date. "Not inspected" is not "not indexed".
12. AI answers: versioned question set (25 questions, US, English, six engines, competitor name variants for Hone, Razer Cortex, EXM Tweaks, NVIDIA App, AMD Adrenalin and MSI Afterburner). "Who AI recommends instead" lists every brand named in the unbranded answers, not only that set. Count each brand once per answer. Mention rate = answers naming Hypertune ÷ answers received. Citation rate = answers linking to hypertune.gg ÷ answers received. Accuracy = answers about Hypertune with no wrong fact ÷ answers about Hypertune. Show coverage out of 150 checks.

6. Where each number comes from

7. Connecting Ahrefs

1. No Ahrefs seat is needed. The platform reads Ahrefs through its API with a key.
2. A separate dashboard key exists (valid to 20 Oct 2027). It is shared only in the private chat with Vin, Austin and Kevin, and stored as a server-side secret, never in Linear or the code repo.
3. Ahrefs Lite has no per-key cap, so the dashboard enforces its own: one weekly cached job with a lock, checked against usage before each pull (the usage check is free). Pause if this key passes 2,000 units in a billing month or the workspace has under 5,000 units left. If usage can't be checked, the pull doesn't run.
4. Charged calls cost at least 50 units, more with more rows and fields, so keep requests small and record actual use.
5. Brand Radar: read only the saved report's custom prompt responses (prompts = custom), once per weekly run, store them, and compute mention rate, citation rate, share of voice and cited sites in SQL. Brand Radar's summary endpoints cost extra units. The report with our 25 questions is being created this week.
6. The AI tools on one-month tests (Otterly, Peec, Scrunch) are not connected.

8. Phases

Fri 30 Oct is the target for Phase 1, subject to Kevin's estimate.