Search visibility, AI discovery, paid efficiency, an external read on what's working, what isn't, and what to act on this cycle. Scoped to US users on roomsketcher.com.
The Q3 2026 Marketing OKRs (June '26) are now the frame for this report: main goal new subscriptions, main activity lead generation. Each objective below is scored against this cycle's data. Initiative status is what we can observe from the outside — owners should correct anything that has moved.
| Key Result | Target | Baseline (Jul 20–Aug 2) | Current (Aug 3–16) | Read |
|---|---|---|---|---|
| PPC ROAS | > 1.0 | 0.67 | 0.78 | +0.11, still short |
| Incr. Market Share (Google-measured) | Increase | 31.8% | 44.5% | Spend-weighted search impr. share* |
*The share gain is real but partly compositional — spend concentrated into Branded (88.6% impression share) while Search-NORAM was throttled to 10.7%. The ROAS gain is the more meaningful move: the account was told new targets this cycle (all campaigns now 0.8, Search 0.7 — see Section 8) and delivered 0.78 combined, up from 0.67. Initiatives: PPC Keyword Expansion (Wida, due Jul — status unconfirmed, Search-NORAM is now budget-capped 57% of the time, which would smother an expansion); Remarketing Setup (Wida, due Aug — Paid Social sessions rose 93→154, consistent with Meta campaigns starting; please confirm); Demand Gen Test (due Sep).
| Key Result | Target | Baseline | Current | Read |
|---|---|---|---|---|
| Non-Brand Clicks (US) | Increase | 917/day | 872/day | −4.8% |
| Non-Brand CTR (US) | Increase | 0.977% | 0.956% | −0.02pp |
| Sign-Up Rate (US) | Increase | 7.01% | 6.13% | −0.88pp — see Section 8 |
Non-brand search is close to flat — a genuine improvement on the −17% of last cycle, and the "creator" keyword family grew 39% in clicks. The sign-up-rate drop is real but composition-driven: the paid pullback removed high-converting clicks, and a new attribution leak (Section 8) is moving signups into Unassigned. Initiatives: Old Blog Article SEO (Trude + Kworq, due Aug); Key Page UX Update (Trude, due Aug); Streamline FPG & Comparisons (Josefine, due Sep); YouTube SEO Creation (Sean, due Sep).
| Key Result | Target | Baseline | Current | Read |
|---|---|---|---|---|
| Established Position in LLMs | Hold / grow | 4 queries 87–96% | 5 queries 75–95% | ChatGPT sweep; mixed elsewhere |
| Website Fully Updated (AI Render) | Sep | — | Not yet measured | Initiatives due Sep |
The commercial signal is strong: ChatGPT referral sessions hit an all-time weekly record (412 in the week of Aug 9) and ChatGPT scored a clean 1.0 on all five tracked queries. The watch items are Gemini/Claude answers dissolving "best overall" into per-use-case picks, and Google AI Mode going unmeasured this cycle. Initiatives: AI Render 1+2 Promotion (Jo, Sep); Legacy Content Update (Trude, Sep); FPG AI Render Update (Josefine, Sep). Enablers: Brand Platform Update, Target Group Overview, Agentic AI Dashboard Test (all Jo, Sep) — the Brand Platform doc matters directly to this KR: it is the context file AI tools will be fed.
Every figure below is filtered to United States users on roomsketcher.com. Session and click metrics are reported as daily averages so windows of different length stay comparable. The Δ column compares Current (Aug 3–16) against Baseline (Jul 20–Aug 2). Both windows are fresh pulls, so baseline figures can differ slightly from what Issue #8 printed for the same dates as conversions matured.
| KPI | Baseline Jul 20 – Aug 2 | Current Aug 3 – Aug 16 | Δ vs Baseline | Status |
|---|---|---|---|---|
| Website Traffic (US) — daily averages | ||||
| Total Website Sessions | 4,341/day | 3,674/day | -15.4% | |
| Direct Sessions | 2,255/day | 1,811/day | -19.7% | |
| Organic Search Sessions | 1,109/day | 1,086/day | -2.1% | |
| Paid Search Sessions (Google) | 462/day | 278/day | -39.9% | |
| Paid Other Sessions (Microsoft) | 236/day | 158/day | -33.0% | |
| Unassigned Sessions new row — tracking leak, see Section 8 | 36/day | 82/day | +129.0% | |
| Referral Sessions | 74/day | 71/day | -4.2% | |
| Email Sessions | 38/day | 34/day | -12.0% | |
| Organic Social Sessions | 33/day | 31/day | -6.9% | |
| Bing & Alt-Engine Organic (US) — daily averages | ||||
| Bing Organic Sessions | 77.9/day | 65.9/day | -15.4% | |
| DuckDuckGo Sessions | 82.9/day | 77.6/day | -6.4% | |
| Yahoo Sessions | 27.4/day | 25.1/day | -8.5% | |
| AI Referral Traffic (US) — daily averages | ||||
| Total AI Referral Sessions | 25.5/day | 50.1/day | +96.4% | |
| ChatGPT | 20.0/day | 46.2/day | +131.1% | |
| Gemini | 2.0/day | 1.9/day | -7.1% | |
| Perplexity | 2.1/day | 1.1/day | -44.8% | |
| Copilot | 0.6/day | 0.4/day | -33.3% | |
| Claude | 0.8/day | 0.4/day | -45.5% | |
| Search Console (US) | ||||
| Clicks Current window has 13 days of GSC data (normal lag) | 996/day | 944/day | -5.2% | |
| Impressions | 94,136/day | 91,549/day | -2.7% | |
| Non-Brand Clicks branded queries excluded — Q3 OBJ 2 KR | 917/day | 872/day | -4.8% | |
| Non-Brand CTR | 0.977% | 0.956% | -0.02pp | |
| Average CTR | 1.059% | 1.032% | -0.03pp | |
| Average Position | 10.27 | 10.92 | +6.3% | |
| Key Head-Term Positions (US) — lower is better | ||||
| “floor plan software” | 8.9 | 9.7 | +9.0% | |
| “floorplan software” | 6.8 | 5.7 | -16.2% | |
| “floor plan app” | 8.7 | 8.2 | -5.7% | |
| “floor plan” | 7.5 | 6.5 | -13.3% | |
| “floor plan creator” | 21.4 | 17.1 | -20.1% | |
| “free floor plan creator” | 9.2 | 4.8 | -47.8% | |
| “floor plan maker” | 24.7 | 25.4 | +2.8% | |
| “room planner” | 20.9 | 23.3 | +11.5% | |
| Template Gallery (/floor-plan-gallery/) | ||||
| Gallery Clicks | 91.2/day | 88.9/day | -2.6% | |
| Gallery Impressions | 12,287/day | 11,912/day | -3.1% | |
| Gallery Pages Receiving Impressions | 843 | 869 | +3.1% | |
| Gallery Pages Receiving Clicks | 172 | 156 | -9.3% | |
| Google Ads — Spend & Efficiency (US, NOK converted at 0.091) | ||||
| Spend | 1,602/day | 1,105/day | -31.0% | |
| Clicks | 926/day | 548/day | -40.8% | |
| CTR | 9.48% | 7.36% | -2.12pp | |
| Total Conversions (all actions) | 157.5/day | 84.8/day | -46.2% | |
| Blended Cost per Conversion | 10.17 | 13.04 | +28.2% | |
| Google Ads — Free Signups (US) | ||||
| Free Account (Personal) | 136.1/day | 69.0/day | -49.3% | |
| Free Account (Business) | 12.1/day | 7.4/day | -38.6% | |
| Google Ads — Paid Conversions (US) | ||||
| New Subscription | 7.87/day | 6.40/day | -18.7% | |
| Cost per New Subscription | 203 | 173 | -14.8% | |
| Premium Project | 1.07/day | 1.00/day | -6.7% | |
| Ready Made Purchase | 0.36/day | 0.93/day | +160.0% | |
| Microsoft (Bing) Ads (US) — NOK converted at 0.091 | ||||
| Spend | 382/day | 260/day | -32.0% | |
| Clicks | 432/day | 270/day | -37.5% | |
| Total Conversions | 33.7/day | 22.9/day | -32.0% | |
| Blended Cost per Conversion | 11.32 | 11.32 | +0.0% | |
| New Subscription | 2.00/day | 1.39/day | -30.4% | |
| Cost per New Subscription | 191 | 186 | -2.6% | |
| Combined Paid (Google + Microsoft, US, USD) | ||||
| Combined ROAS conversion value ÷ spend — Q3 OBJ 1 KR target > 1.0 | 0.67 | 0.78 | +16.4% | |
| Google Ads ROAS (US) | 0.67 | 0.80 | +19.4% | |
| Google Ads ROAS incl. renewals | 0.81 | 0.91 | +12.3% | |
| Microsoft Ads ROAS (US) | 0.69 | 0.69 | 0.0% | |
| Spend-weighted Search Impr. Share rises partly because spend concentrated into Branded | 31.8% | 44.5% | +12.7pp | |
| Combined Spend | 1,984/day | 1,365/day | -31.2% | |
| Combined Conversions | 191.3/day | 107.7/day | -43.7% | |
| Combined Cost per Conversion | 10.37 | 12.68 | +22.3% | |
| Combined New Subscriptions | 9.87/day | 7.79/day | -21.1% | |
| Combined Cost per New Subscription | 201 | 175 | -12.9% | |
| GA4 Conversion Events (All Sources, US) | ||||
| Sign-Up Rate sign_up ÷ sessions, US — Q3 OBJ 2 KR | 7.01% | 6.13% | -0.88pp | |
| sign_up | 304.4/day | 225.1/day | -26.0% | |
| begin_checkout | 54.9/day | 53.0/day | -3.5% | |
| new_subscription | 16.6/day | 15.3/day | -7.8% | |
| purchase | 30.9/day | 29.4/day | -4.6% | |
| Site Performance (local Lighthouse — same method as Issue #8, so deltas are meaningful; not PSI-comparable) | ||||
| Mobile Performance (median of 4) | 84 | 92 | +8 pts | |
| Mobile LCP (median of 4) | 4.01s | 2.96s | -26.2% | |
| Desktop Performance (median of 2) | 94 | 95 | +1 pt | |
| Desktop LCP (median of 2) | 1.48s | 1.37s | -7.4% | |
| Accessibility | 97 | 97 | — | |
| SEO | 100 | 100 | — | |
sign_up fell 26%, but paid-attributed signups fell with the deliberate spend cut, Direct signups fell with paid's echo, and a new attribution leak moved signups into Unassigned (40→276) as Paid Search signups halved (482→213). Intent held: begin_checkout is down just 3.5% and purchases 4.6%.Channel mix for the current window, August 3 – 16, 2026. 51,433 total US sessions — down 15.4% on the baseline, with almost all of the loss in the two paid channels and the Direct traffic that moves with them.
Issue #8's headline was a five-month organic decline bottoming at 7,318 weekly sessions — the lowest on record. This cycle the line stopped falling: two consecutive weeks of recovery off that floor, while the head terms kept improving. Two weeks is stabilization, not a trend — but the “uninterrupted slide” framing no longer holds.
Total US sessions fell 667/day against the baseline. Decomposed by channel, the story is not subtle:
Why this matters: a 15% traffic drop reads as an emergency; a 31% deliberate spend cut whose session cost was concentrated in low-value clicks reads as a trade. The scorecard's efficiency rows (cost per subscription −13%, ROAS +0.11) say the trade has so far been favorable on the Q3 KR. The open question is whether the subscription volume loss (−21% combined) is acceptable while it lasts.
Issue #8 flagged the Direct/paid-spend correlation as “suggestive, not conclusive” and asked for another cycle of evidence. This cycle provided it: spend stayed cut through August 3–12 and Direct held at the depressed level rather than reverting — weekday Direct averaged ~1,800/day against ~2,380/day in mid-July. The practical implication stands: paid's true contribution is understated by its own reporting, and any budget decision should price in the Direct echo. A clean brand-holdout test remains the way to size it properly.
AI referral traffic nearly doubled this cycle — 357→701 sessions, the best 14-day window since tracking began — and ChatGPT scored a clean 1.0 on all five tracked visibility queries. Last issue's central worry (recommended but not clicked) inverted: the recommendation is now shipping clicks at record volume.
| Engine | Baseline Jul 20 – Aug 2 | Current Aug 3 – 16 | Δ vs Baseline |
|---|---|---|---|
| ChatGPT (chatgpt.com + openai) | 280 | 647 | +131.1% |
| Gemini (gemini.google.com) | 28 | 26 | −7.1% |
| Perplexity (both sources) | 29 | 16 | −44.8% |
| Claude (claude.ai) | 11 | 6 | −45.5% |
| Copilot (copilot.com + copilot.microsoft.com) | 9 | 6 | −33.3% |
| Total AI Referral Sessions | 357 | 701 | +96.4% |
perplexity.ai and perplexity. bing and cn.bing.com are counted as Bing organic, not AI.ChatGPT sent 647 US sessions this window, up 131% — and the week of August 9 (412 sessions) is the single best AI referral week since measurement began, beating the previous May peak. This is not a blip in a noisy series: the surge is one engine, sustained across the window, and it coincides with ChatGPT naming RoomSketcher “Best overall” on every one of the five visibility queries captured August 17.
sign_up at 5.1% (35 signups on 688 sessions) — 2.4× organic search. The assistant answers “which tool”; the visit is confirmation.Concentration deepened: every non-ChatGPT engine declined this window, and Perplexity's referrals nearly halved (29→16) — consistent with its captures this cycle, which twice demoted RoomSketcher from “best overall” to first-among-unranked-options. The volumes are too small to act on individually, but the pattern matters: the AI channel's growth is one engine's editorial behavior, and it can change without notice. The hedge is the same as it was — be the citable source on the pages assistants ground on, not just the name they remember.
Five queries this cycle: the two locked anchors, the two spelling variants, and a new natural-language query — “im looking for software to make floor plans on my apartments” — added because buyers increasingly ask assistants in sentences, not keywords. Each cell is scored 0–1: 1.0 = definite #1 / best-overall, 0.75 = wins a named sub-category while another tool or nothing holds the overall slot, 0.5 = listed among the recommendations, 0.25 = mentioned but not recommended, 0 = absent.
The capture set covered ChatGPT, Gemini, Perplexity, Copilot and Claude but not Google AI Mode (which requires three runs per query under the method locked last issue). To keep deltas honest, this issue re-computes the August 3 reading on the same five engines and compares like-for-like. AI Mode returns to the set next cycle; the locked anchor series in history.csv records this reading as 5-engine.
| Engine | “floorplan software” Aug 3 → Aug 17 | “floor plan creator” Aug 3 → Aug 17 | NL “apartments” new | What the answer said this reading |
|---|---|---|---|---|
| ChatGPT | 1.0 → 1.0 | 1.0 → 1.0 | 1.0 | “⭐ Best overall” table lead on every query; NL answer opens “1. RoomSketcher — best overall.” |
| Perplexity | 1.0 → 1.0 | 1.0 → 1.0 | 0.75 | Creator: “widely praised as the best overall.” NL: co-lead — #1 for precise/polished plans, Floorplanner holds the quick-browser slot. |
| Copilot | 1.0 → 1.0 | 1.0 → 1.0 | 1.0 | #1 in “🏆 Best Overall Picks” (software) and “the best all-around pick for most people” (creator); NL Top Picks #1. |
| Gemini | 0.75 → 0.75 | 1.0 → 0.75 | 1.0 | Anchors: owns “Best for Real Estate & Interior Designers,” no overall winner named. NL: “Best Overall for Beginners: RoomSketcher,” first in table. |
| Claude | 0.75 → 1.0 | 0.75 → 0.5 | 1.0 | Software: “gets the top overall pick from several roundups.” Creator: shared go-to with Floorplanner and discounts the #1 claim as vendor-sourced. NL: “generally considered the best overall.” |
| Total (5-engine) | 4.50 → 4.75 | 4.75 → 4.25 | 4.75 | 95% / 85% / 95% |
On the “best floor plan creator” anchor, Claude recommended RoomSketcher for real-estate plans — then added, unprompted: “though take that ranking with a grain of salt since it comes from RoomSketcher's own blog.” Its software answer generalized the point: “a lot of 'best floor plan software' articles are written by the vendors themselves.”
The new query — phrased the way a landlord actually types — scored 95%: ChatGPT, Gemini, Copilot and Claude all named RoomSketcher the top pick, several explicitly for rental-listing use. Perplexity made it co-lead. This matters more than the keyword anchors: it is closest to real buyer behavior, and it lands squarely on the real-estate/landlord positioning every engine already concedes to RoomSketcher. Baseline set; tracked from this issue in history-variants.csv.
The creator-family bet is starting to pay: “floor plan creator” improved four positions, the free-variant queries broke into the top 5, and family clicks rose 39%. The uncovered terms — “maker” and the AI phrasings — are unchanged and still unclaimed.
The creator keyword family grew from 100 to 139 clicks (impressions 2,613→3,315) — the strongest family on the board. The composition is telling:
/floor-plan-creator/ page itself is still grinding: impressions 29→41/day and position 37.3→32.6, but only 1.8 clicks/day. Google is widening its testing without yet committing to a page-1 slot. Normal for month two; the page's rank — not the family's — is the number to watch next cycle.Comparison-intent search is warming up exactly where the LLM work is aimed: the query “best floor plan software” produced 13 clicks on 330 impressions this window against 2 on 269 in the baseline, with the comparison blog post capturing them. Small numbers, but this is the query family that feeds both the blue-link comparison shopper and the LLM grounding pipeline — the same post Claude cited (and then discounted as first-party). Third-party corroboration would compound here first.
Re-verified August 17: /floor-plan-maker/, /floor-plan-app/ and /ai-floor-plan-generator/ all return 404. “Floor plan maker” sits at position 25.4 on 21 impressions/day with no dedicated page, while floorplanmaker.ai holds the exact-match domain, and the AI-native cohort continues to organize around the generator phrasing. The creator page's trajectory is the proof this trade works; the maker page remains the cheapest unclaimed replica of it.
Spot-checks this cycle: Planner 5D has gone fully AI-first in its positioning, Cedreo and Floorplanner remain AI-silent — and Homestyler, an AI-heavy consumer tool, took a “Best overall” slot from RoomSketcher in one of Copilot's answers. The incumbent set is stable; the AI framing inside it is being claimed.
| Competitor | “Creator” Page | “Maker” Page | AI Positioning | Positioning / What Changed This Cycle |
|---|---|---|---|---|
| RoomSketcher | Yes | No (404) | Features, not framing | Professional results, beginner-friendly; real-estate/listing workflows. AI Render launch is the Q3 vehicle for claiming the AI frame. |
| Planner 5D | Yes | Yes | AI-first | Homepage now leads with “all-in-one AI-powered home design software” — AI Studio, AI floor-plan recognition, AI 3D rendering, plus Apple Vision Pro. The most aggressive AI repositioning among incumbents. |
| Homestyler | — | — | AI-heavy | Newly visible in LLM answers: Copilot named it “Best overall” on the two-word software query this cycle, relegating RoomSketcher to “easiest to use.” Worth tracking as an answer-surface rival, not just a product rival. |
| Cedreo | Yes | — | None | Unchanged: builders/contractors, “complete house in 2 hours,” no AI branding. Not contesting the homeowner end. |
| Floorplanner | Yes | Yes | None | Homepage still leads with heritage (“since 2007”) and free access; no AI messaging. Remains the most frequent co-recommendation next to RoomSketcher in LLM answers. |
| SmartDraw | Yes | Yes | Partial | Diagramming breadth; strong exact-match SEO footprint. No change observed. |
| SketchUp | — | — | Partial | Professional 3D; appears in nearly every LLM answer as the “advanced users” pick. |
| floorplanmaker.ai / Maket.ai / Snaptrude | — | Exact-match domain | Native | AI-native cohort unchanged this cycle; still building on “maker” and “AI generator” vocabulary RoomSketcher has no page for. |
On “best floor plan software” (the two-word form — 12.7× the search volume of the one-word anchor), Copilot's ranked list read: 1. Homestyler — Best overall; 2. ArchiCAD — Best value; 3. RoomSketcher — Easiest to use. That is the first time this series has recorded a consumer AI-design tool displacing RoomSketcher from an overall-best slot. One engine, one query, one reading — but Homestyler's whole pitch is AI-generated design, and Copilot's grounding (worldmetrics.org-style stat roundups) rewards exactly the kind of third-party content Homestyler is seeding. Watch whether it recurs next reading before responding.
Planner 5D's homepage now opens with “all-in-one AI-powered home design software” — the positioning RoomSketcher's Q3 Objective 3 exists to contest. Cedreo and Floorplanner have left the frame unclaimed, so within the incumbent set this is currently a two-horse race with Planner 5D ahead on messaging (not on capability). The Objective 3 initiatives (AI Render promotion, legacy 3D-photo replacement, FPG AI Render refresh with alt tags) are the right instruments; the competitive read says September delivery matters — every cycle the frame is uncontested, round-up authors and LLM training data settle further on “Planner 5D = the AI one.”
Four things went right this cycle, and two of them are the direct payoff of decisions made at the last call. Worth naming so they don't get changed by accident.
Cost per new subscription fell $201→$175 (−13%), combined ROAS rose 0.67→0.78, and Google's ROAS including renewals reached 0.91 — within sight of the 1.0 KR for the first time. The spend that was cut was buying free accounts at scale (Google's free-account conversions halved) while subscriptions held far better (−19%). This is what value-based discipline looks like in the numbers; the remaining question is volume appetite, not efficiency.
An all-time-record 647 sessions at a 5.1% signup rate, backed by a clean 1.0 sweep on all five visibility queries. Two years of “AI is coming” talk has resolved into a channel that outperforms Email and Organic Social combined. The LLM position driving it is an asset to defend — which is what the third-party-corroboration work in Section 10 is for.
After five months of decline, organic sessions were flat (−2.1%) with two consecutive weeks of recovery off the July floor. Six of nine tracked head terms improved, “floor plan” itself reached position 6.5, and the creator family grew clicks 39%. The domain's authority on its core vocabulary kept compounding straight through the traffic trough — the decline was never an SEO-quality problem, and the recovery is arriving with rankings intact.
Pages earning impressions rose 843→869 while clicks held within 3% of baseline through a soft-traffic window. The gallery remains the most decline-resistant asset on the property, which is exactly the property Q3's FPG initiatives (streamlined production, AI Render imagery, alt tags) are designed to scale. No change of course needed — just delivery.
PSI was quota-blocked for the third consecutive cycle, so Lighthouse ran locally again — same machine and method as Issue #8, which makes this the first cycle where the deltas are meaningful. They are all positive.
Same method, same machine, two weeks apart: mobile LCP median went 4.01s→2.96s and the worst-of-four run went from 10.85s to 5.23s. The bimodal “fast page or 3×-slower page” behavior Issue #8 flagged has narrowed markedly. Local runs are noisy, so treat the direction as real and the magnitude as approximate — but a halving of worst-case LCP does not usually happen by chance.
wpmedia.roomsketcher.com, or unused-JS cleanup), it worked and should be noted so it doesn't regress silently.The keyed and unkeyed PSI endpoints both returned daily-quota errors again. Two audit runs cannot consume 25,000 queries/day, so something else is exhausting the shared API project. The trend line against Issues #1–#7 is now nine-plus weeks stale and cannot be rebuilt from local runs. The fix is administrative, not technical: create a fresh Google Cloud project with its own PageSpeed API key for audit use. Ten minutes of setup ends a three-cycle data gap.
One new critical item leads: a tracking regression that appeared almost exactly at the window boundary and is quietly rewriting channel attribution. The paid items that follow are questions of intent — the account moved this cycle, but not quite in the direction the KR points.
Three symptoms with the same timestamp and shape:
sign_up events went 40 → 276 between windows, while Paid Search sign_up fell 482 → 213. The paid decline is partly real (spend was cut) — but the mirrored rise in Unassigned means a chunk of paid-driven signups are now arriving without attribution.Likely candidates, in order: a GTM container change (tags unpublished or consent-mode default flipped), Google Ads auto-tagging/gclid loss on a redirect, or a landing-page change stripping URL parameters. This sits on top of the standing issue — 80% of new_subscription events (171 of 214) still land in Unassigned — but it is new, it is growing, and it will corrupt every channel comparison in this report until found. Kworq action #3.
Issue #8 recommended raising every target ROAS toward the 1.0 KR. The targets did change this cycle, in a way that reads as normalization rather than ambition:
| Campaign | Target (Aug 3) | Target (now) | Actual ROAS (this window) | Read |
|---|---|---|---|---|
| 0. S – Branded | 0.94 | 0.80 | 1.10 | Lowered while over-delivering |
| 1. S – Search | 0.80 | 0.70 | 0.44 | Lowered; still worst in account |
| 2. S – Competitors | 0.74 | 0.80 | 0.64 | Raised — right direction |
| 3. P-Max | 0.74 | 0.80 | 0.65 | Raised — right direction |
| Combined US (incl. Microsoft) | — | — | 0.78 | KR needs > 1.0 |
Raising Competitors and P-Max to 0.8 is the staged move we suggested. But lowering Branded to 0.8 while it returns 1.10 licenses Smart Bidding to pay more for brand clicks that were already profitable — and Branded's spend did rise 20% this window at a ROAS that fell from 1.22 to 1.10. If the lowered targets were part of a planned staircase, the next step up should be scheduled; if they weren't deliberate, Branded's should go back up first.
The campaign that was cut ~85% in late July now runs a configured ~6,600 NOK/day ($600) but loses 57% of its eligible impressions to budget — the auction wants far more than it is allowed to spend. Meanwhile its ROAS improved from 0.31 to 0.44 post-cut: the spend that remains is better spend. Two coherent choices exist — fund it (especially if the July keyword expansion added new AI-feature terms that have never had room to run) or cap it deliberately and say so. The current state, where the OBJ 1 “broader reach” initiative and a 57% budget ceiling coexist on the same campaign, is the one configuration that can't be right.
Account CTR dropped 9.48%→7.36%, which looks like relevance decay until it is decomposed: with Search-NORAM throttled, the Competitors campaign is now 51% of remaining impressions (53,170 of 104,286) at its native ~4.5% CTR, while high-CTR Branded and Search shrank in the mix. Per-campaign CTRs are broadly stable. No action needed beyond knowing the aggregate is no longer comparable across the budget change.
The external frame shifted in RoomSketcher's favor this cycle: assistants are not just answering — they are increasingly sending the click, and the brands they recommend collect it.
The 2025 story was AI answers absorbing clicks. This cycle's data shows the sequel: when an assistant commits to a single recommendation with a linked citation, it concentrates the surviving clicks on that one brand. RoomSketcher's ChatGPT referrals doubling while its blue-link non-brand clicks held flat is that mechanism in miniature. The strategic consequence: LLM position is no longer just brand influence — it has direct, measurable session and signup yield, which means GEO work can now be evaluated like a channel rather than like PR.
Claude's unprompted “vendor-owned content” caveat this cycle is part of a broader pattern: engines increasingly weight independent review platforms, editorial round-ups and forum consensus over first-party claims when adjudicating “best.” For every brand that built its answer-surface presence on its own comparison content — RoomSketcher included — the moat is migrating from owning the ranking page to being corroborated by pages nobody owns. That is a content-strategy instruction with a deadline measured in model-training cycles.
Seven items for the next 14 days, ordered so the confirmations and diagnostics that unblock everything else come first. Items marked KWORQ are ours to execute. Objective tags map each action to the Q3 OKR it serves.
All four campaigns' targets changed since the last call (now 0.8 across the board, Search at 0.7). Raising Competitors and P-Max was the staged move toward the 1.0 KR; lowering Branded from 0.94 to 0.8 while it returns 1.10 points the other way. We need to know whether this was a planned staircase or a normalization — and if the former, the next increment should go on the calendar now, since Smart Bidding needs a learning window per step.
The campaign's economics improved after the July cut (ROAS 0.31→0.44) but it now runs into its ~$600/day budget ceiling most of the day. If the July keyword-expansion initiative added terms to this campaign, they have never had room to prove themselves. Either fund it in steps while watching ROAS, or cap it as policy — both are defensible; the accidental version isn't.
Unassigned signups went 40→276, “(not set)” sessions nearly tripled, and Paid Search signups halved beyond what the spend cut explains — all starting at the window boundary. We will audit the GTM container's change history, auto-tagging/gclid pass-through, and consent-mode state, and restore attribution. This also covers the long-standing 80%-Unassigned new_subscription issue in the same pass, since the instrumentation is shared.
This cycle's reading covered five engines; Google AI Mode (three runs per query, median-scored) returns next cycle so the locked anchors regain their full series. The new natural-language apartments query is now a permanent fifth tracked query. Alongside it: pull the Bing Webmaster Tools AI-citations CSV, now missing for three cycles — it is the only view of grounding volume as distinct from clicks.
Carried forward, with stronger evidence: the creator page's family grew clicks 39% this cycle and the free-variant queries broke into the top 5. “Floor plan maker” still has 21 impressions/day of standing demand at position 25 with no page behind it, and floorplanmaker.ai still holds the exact-match domain.
Claude now discounts RoomSketcher's own #1 ranking as vendor-sourced, while citing Capterra and The Close as neutral arbiters. The counter is presence in sources nobody owns: refreshed review-platform profiles, pitches to the independent round-ups the engines actually cite, and authentic participation where floor-plan questions get asked. This is the highest-leverage Objective 3 work that isn't already an initiative.
Carried forward: /ai-floor-plan-generator/ is still a 404 while the AI-native cohort organizes around the phrase and Planner 5D claims the adjacent “AI home design” frame. With AI Render shipping, the product substance now exists. The decision needed is whether to claim the query family or explicitly cede it; the build can follow the decision.
| Term | Meaning |
|---|---|
| SEO | Search Engine Optimisation — earning visibility in conventional search results. |
| AEO | Answer Engine Optimisation — being the source an AI assistant cites when it answers directly. |
| GEO | Generative Engine Optimisation — shaping how generative AI systems represent and recommend a brand. |
| GSC | Google Search Console — source for impressions, clicks, CTR and average position. |
| ROAS | Return on Ad Spend — conversion value divided by spend. The Q3 Objective 1 KR is ROAS > 1.0. |
| LCP | Largest Contentful Paint — time until the largest visible element renders. Google's "good" threshold is 2.5s. |
| CLS | Cumulative Layout Shift — how much the page moves while loading. Below 0.1 is "good". |
| TBT | Total Blocking Time — how long the page is unresponsive to input during load. |
| CTR | Click-Through Rate — clicks divided by impressions. |
| CPC | Cost per Click. |
| CPA | Cost per Acquisition — spend divided by conversions. In this report, "blended" CPA counts all conversion actions including free accounts. |
| P-Max | Performance Max — Google's automated campaign type spanning Search, Display, YouTube and Shopping inventory. |
| Unassigned | GA4's bucket for sessions and conversions where the traffic source could not be determined — see Section 8. |