#build-in-public #agents #seo
FurnishDesk, month one: 778 impressions, 2 clicks, 1 RFQ
On 17 September my partner Shishir and I put FurnishDesk on the internet. Twenty-three days later the whole scoreboard fits in one line: 778 Google impressions, 2 clicks, 1 RFQ.
Nobody frames a launch screenshot like that. It’s what happened, though, so here’s month one.
What it is
FurnishDesk sources commercial furniture for offices, hotels, restaurants, cafeterias, clinics and schools. A buyer sends one requirement, say workstations for an office floor plus the chairs and storage that go with them, and gets back one consolidated quotation. The desk earns a margin on it.
It isn’t a manufacturer and it holds no inventory. The pitch is that a buyer deals with one desk instead of chasing several vendors and then trying to compare quotes that were never written to be compared. That’s the idea, anyway. One month in, nothing has proved it yet.
Shishir runs the desk. The back office is two AI agents running inside Agent OS, the system I run my companies on:
- shishir-procurement looks after the server, the RFQ intake and the CRM.
- furnishdesk-content writes the SEO pages and tracks how each one ranks, week by week.
Both act through the same governed gateway as every other agent I run. The daily grind of “write the Nagpur page, check where Surat sits this week” is theirs.
What launched
About twenty pages. Category pages, plus an “office furniture in X” page for a set of tier-2 cities: Jaipur, Surat, Vadodara, Nagpur, Indore, Raipur, Bhopal, Ahmedabad. A few metro and national pages went up alongside them.
The tier-2 bet was deliberate. A new domain with zero links seemed unlikely to get anywhere for “office furniture India”, and the working theory was that smaller cities have less competition, so a decent page might have a chance there.
The numbers
| First 23 days | |
|---|---|
| Google impressions | 778 |
| Clicks | 2 |
| RFQs | 1 |
| Average position, tier-2 city pages | about 19 to 23 |
| Average position, metro and national pages | 48 to 63 |
Don’t read the first three rows as a funnel. I’m not claiming the RFQ came from either of those clicks.
The positions are averages across every query a page showed up for, so they aren’t a rank you could screenshot. Still, the gap is wide. The tier-2 pages sit somewhere around pages two and three, and the metro and national pages are a lot further back. That fits the small-city theory. It doesn’t prove it (the pages differ in age, content and which queries they catch), and none of it has turned into enquiries yet.
2 clicks from 778 impressions is a click-through rate of about a quarter of a percent. The first clicks came in through the Jaipur page.
On 1 October a batch of pages got rewritten, and daily impressions went from about 24 to about 48 after that. I would love to say the rewrites did it. With three weeks of data on a domain Google has barely met, I can’t separate “the rewrites worked” from “Google is warming up to a new site and would have doubled anyway”. I’m logging it as a coincidence I’m happy about.
Lesson 1: the enquiry nobody saw for six days
The one real RFQ arrived and sat there for six days.
The form worked. The lead was stored and showed up on the admin page exactly as designed. What was missing was any step where a person got told, and that gap wasn’t even hidden. The runbook said it in plain words: “No notification yet. New RFQs are only visible in /admin; someone has to check it.”
So it was documented, and nobody owned the checking. For six days nobody checked.
Six days is a long time to leave someone waiting on a quote they asked for. Every RFQ now fires a Telegram alert, which took less time to build than the delay lasted.
What bothers me is where my attention went: all of it on whether the agents could build the intake, none of it on whether a human would hear about what came through.
Lesson 2: Google found the pages and didn’t crawl them
Five or six of the newer pages sat in Search Console as discovered but not indexed. Google’s own description of that status is “The page was found by Google, but not crawled yet.”
One suspect was self-inflicted. Package links on the site carried a ?package= parameter, which produced 52 near-duplicate URLs on a site with about twenty real pages. Those links are now nofollowed and the parameter is blocked in robots.txt.
I don’t know that the duplicates caused the stall. Counting URLs doesn’t tell you which ones Google actually fetched. And while checking this post I learned the robots.txt block is the blunt option. Google’s duplicate URL guide says “Don’t use the robots.txt file for canonicalization purposes” and recommends redirects or a canonical tag instead, and a blocked page can’t show Google the canonical tag on it. That’s going back to the content agent as a question.
The other suspect is that nothing on the internet links to furnishdesk.in.
That’s the other reason this post exists, and I’d rather say so than pretend otherwise. Pages like cafeteria furniture are among the ones still waiting to be crawled. If they get indexed I’ll say so, but a robots.txt change and a first link landing in the same month won’t tell me which one did it.
Lesson 3: the top of the page isn’t websites
For “office furniture in [city]”, the top of Google is directories, IndiaMART and JustDial, and then the Maps pack. Individual sites come below those. The sellers listed in the directories are still competing for the same buyer. They just reach that buyer through a different door than a standalone site does.
The other surprise came from search volumes. In every city we measured, people search for office chairs as much as, or more than, for office furniture. That looks like the obvious page to build next.
I’m not convinced it is. My guess, and it is a guess, is that most chair searches are retail: one person buying one chair. A desk built to consolidate a whole fit-out into one quote doesn’t want that traffic. But a search volume can’t tell you who is searching, and businesses buy chairs on their own too. The only thing that would settle it is enquiries, and I have one.
What I’m watching
Twenty-odd pages, two agents, one partner doing the part that needs a human, and one RFQ that should have been answered on day one instead of day six.
Two things for month two. Do the stuck pages get indexed? And do any of the tier-2 pages reach page one, where clicks actually happen?