Weekly organic leads rose from 6 in early July to 20 in the week of 24 August 2026. August paid spend was $0.
First-party case study. August record from the brief of 31 August 2026, with Google Search Console complete through 28 August. Latest update reported 21 September 2026.
First-party case study · August 2026
In June, Ruh AI spent $9,001 on Google Ads. In August it spent nothing. Over the same period organic leads went from six a week to twenty, and ChatGPT became the second largest source of them. This is a first-party case study: we did it to ourselves, and every number below names the instrument it came from.
20/wk
organic leads in the week of 24 August, a series record, from 6 a week in early July
PostHog lead events, test entries excluded
0
paid spend in August, down from $9,001 in June. No paid-attributed lead since the week of 3 August
Google Ads and gclid attribution
$23,148/mo
ad-equivalent value, priced at the $643 cost per lead measured from our own June spend
internal ledger
24%
of the last 30 days of leads arrived from ChatGPT, our number two source after Google
PostHog referrer census, n approx 46
Latest update · 21 September 2026
Figures the Ruh AI team reported at its stand-up, before September leads were counted.
How to read these figures1 to 2a day
organic leads, the running rate the team reported
A daily rate, so it is not plotted on the weekly series.
about22%
more construction-focused clicks on Google
The report does not state the comparison period.
Overall traffic is falling on purpose as the site moves away from generic AI content. The team also reported Bing impressions and clicks growing week over week, without figures. The stated aim is relevant construction visitors rather than volume.
Source: Ruh AI marketing stand-up notes. Team-reported, not instrument readings with a stated window.
The timeline
Not a campaign. A system brought online in stages, with checkable dates. The citation chart inflects on one.
Read the platform guide
The drip starts shipping pages continuously. Bing's AI answers had never cited the site before this date. A flat zero on the chart.
First strategy brief on 23 July. From 4 August every action ships against a frozen baseline, so each change can be measured net of the site's own drift.
963 citations a day across 104 pages. 20 organic leads in the week of 24 August. Paid spend nil since early in the month. Automated publishing was switched off on 18 August, with manual pushes since.
The record: July and August
Ads were paused in early July on purpose, to find out whether organic could carry the pipeline on its own. The weekly series answered it: the week of 24 August produced twenty organic leads from sixteen distinct companies, a series record. Across August, 48 form submissions from 36 unique companies.
Priced honestly, that is $23,148 a month of ad-equivalent value, using the $643 cost per lead we measured from our own June spend rather than an index average. Over 90 days: 99 leads and 34 demo bookings.
Organic leads per week
Six recorded weeks, July to August 2026
One purpose-built page on ruh.ai produced 36 enquiries in 90 days while the contact page produced 31 and every blog post produced none.
ruh.ai, captured for the August 2026 study.The citation layer
A flat zero on the chart. That is the week the automated publishing pipeline went live.
Explore AI SearchBing AI citations per day
April to 25 August 2026, readings in order, not to time scaleWhat it showsThe publishing pipeline shipped pages automatically from 13 June until 18 August, with manual pushes since. On 25 August, Bing's AI answers cited 104 of the site's pages 963 times a day.
How it was readDrawn from the readings in Bing Webmaster Tools, operator-read 31 August 2026. Within August the multiple is peak to peak: 292 to 963 citations, 3.3x, and 50 to 104 pages, 2.1x. We do not compute growth from the 1 August reading of 32, which was a Saturday trough.
Roughly 24% of the last 30 days of leads arrived from chatgpt.com, behind Google and ahead of direct. Alongside the human traffic, 1,417 AI-agent pageviews in 90 days against 16,248 human ones.
Selected referrers shown: 42 leads. Bar widths use this subtotal. The roughly 24% ChatGPT figure uses the full census of about 46 leads, not this chart's subtotal.
The record shows average position 6.1 on Bing against 19.5 on Google, alongside ChatGPT referral traffic. These averages use different search populations and are not a ratio of performance. They do not establish which source an AI system used or why a visitor arrived.
| Bing, five weeks | 27 Jul | 17 Aug | Change |
|---|---|---|---|
| Impressions per week | 7,227 | 13,079 | +81% |
| Clicks per week | 300 | 711 | +137% |
| Keywords ranking | 970 | 1,484 | +53% |
| Average position | 6.2 | 6.1 | held while volume grew |
August totals to 28 Aug: 44,237 Bing impressions, 2,102 clicks, 1,389 keywords in Bing's top 10. Tracking began 27 July, so this is five weeks and we do not claim longer.
14 of 18 checked keywords confirmed cited, including the ROM cluster, commercial glazing estimating software, and catch basin in construction. Hand-verified rather than taken from bulk data, because the bulk data was stale. Links in an AI Overview share the position of that Overview, not automatically position one.
14/18
keywords confirmed cited in Google AI Overviews
The method
Each rewrite is compared with the whole site's movement over the same window.
What this does not prove83
rewrites measured against a frozen baseline
53
of them net-positive, 64%
+2.06
mean positions gained, net of the site's own drift
106
keywords newly visible, and +56,163 impressions
NET is the page's move minus the site's move over the same window.
Position before and after, last 28 days. Lower is better.
A page can rise while the whole site rises. Comparing its movement with site-wide drift adds context, but does not isolate the effect of an intervention. These are observational comparisons, not controlled experiments or guaranteed outcomes.
The supporting layer
Named-query Google Search Console figures, and the infrastructure that keeps every page measured.
How the SEO module works| Named-query GSC | July | August to 28th | Change |
|---|---|---|---|
| Keywords ranking | 8,532 | 9,025 | +6% |
| Keywords with real traction, 10+ impressions in 28d | 986 | 1,442 | +46% |
| Impressions, named queries | 73,892 | 83,480 | +13% |
| Keyword churn over 28d | Not recorded | +5,998 gained / 4,896 lost | net +1,102 |
Junk queries excluded, full days only. 3,500 keywords sit in Google's top 10, 39% of those tracked.
691 pages, crawled daily, every one of them measured.
ruh.ai as captured for the August 2026 study.What this does not prove
A case study where everything worked is not believable. These are the things a sceptical reader would find, so we put them here first.
Instruments: Google Search Console, Bing Webmaster Tools, PostHog, Strapi form submissions, and the keyword tracker's receipts ledger, plus the Ruh AI team's stand-up report for the 21 September update. August record from the brief of 31 August 2026, with Google Search Console complete through 28 August. Latest update reported 21 September 2026.
Invite only. We ran this on ourselves before we sold it to anyone. If you want the same instruments pointed at your site, that starts with an audit, not a contract.
