Your phone visitors message you. Your laptop visitors do not. SafaiKaro is a Karachi pest control business that we own and run ourselves, so every number here is ours to check. In the 28 days to 6 August 2026 it took 3 leads from 81 desktop visitors, against 43 from 164 on mobile.
The pages were identical on both. The chat link was not. Local SEO for a small business means getting found by people searching near you. Here it took two jobs at once: publish the pages they search for, then make the message button work on whatever they are holding.
| Measure | Before | After | Measurement window |
|---|---|---|---|
| Lead events per day | 1.38 | 2.23 | 37 days to 23 Jul 2026, then the 39 days after |
| Lead clicks per Google search pageview | 17.3% | 21.5% | the same two windows |
| Pages earning search impressions | 24 | 40 | 35 days each side of the 23 Jul 2026 release |
Every job here starts in a chat window, not a form
SafaiKaro sells fumigation, termite, bed bug and rodent treatment to Karachi households. Almost nobody fills in a form. In that same 28 day window, 39 of the 46 people who made contact did it through WhatsApp. It books about ten paid jobs a month, a figure its founder gave on 10 September 2026.
Autonomous Technologies is the team behind this site. We work with Shopify founders losing orders and hours to systems that do not talk to each other. SafaiKaro is our own venture, so we show you what failed as well as what moved. Client stores get the same work through our growth and measurement service.
The pages people searched for did not exist
In April 2026 the site had 11 indexable pages. An indexable page is one you have told Google it may list. Seventeen thin neighbourhood pages had already been refused as crawled, currently not indexed. The biggest commercial phrase in the market, fumigation services in karachi, appeared nowhere on the site.
Karachi also does not search in English. People type khatmal for bed bugs, deemak for termites and lal beg for cockroaches. Only deemak had a post; khatmal and lal beg were nowhere on the site.
Then the device problem. A wa.me link is a deep link, meaning a link that opens an app. On a phone that is one tap into a chat. On a laptop it lands on WhatsApp Web, useless unless the visitor has already paired their phone.
Split your conversion by device before you rewrite a single page. The likeliest cause of the desktop gap here was not the copy, the offer or the layout. It was one link that assumed a phone. No conversion checklist contains that.
What we built, and who had to approve each piece
The pages came first. Service pages, area hubs, 12 researched district pages and 11 blog posts, most in Roman Urdu or Urdu. Old addresses stayed alive as redirects. An automated job rebuilds the sitemap on every change, so new pages cannot go missing.
A second release, PR #18 on 17 September 2026, added a contract page, an apartment fumigation page, a commercial kitchen cleaning page, a post-construction cleaning page and two commercial cleaning guides. The sitemap holds 52 indexable pages today, up from 43 on 6 September. Results are not measured yet; the next scheduled read is early October.
The desktop card came second. On screens 1,024 pixels and wider with a mouse, any WhatsApp button now opens a small card. It offers three ways out: a code you scan to open the chat on your phone, a number you can copy, and a WhatsApp Web link. Phones keep the behaviour that already worked. It is 139 lines of JavaScript, added to each page with one script tag, and it changed no button markup.
The measurement came third. Every lead click now carries the page it came from and where the button sat. Prices come from one shared file. The chat buttons on the starting-price cards name the service, never a price, and the price finder’s message reads its figure from that same file.
What moved, what stayed flat, and what we never measured
| Measure | Before | After | Measurement window |
|---|---|---|---|
| People clicking to chat or book, per day | 1.19 | 1.59 | 37 days to 23 Jul 2026, then the 39 days after |
| Lead clicks per Google search pageview | 17.3% | 21.5% | the same two windows |
| Price page: search clicks, then lead clicks | 36 and 4 | 27 and 12 | clicks: 35 days each side of 23 Jul 2026; lead clicks: the same two windows |
| Sitewide click-through rate from search | 0.97% | 0.85% | 35 days each side of the 23 Jul 2026 release |
| Referring domains, other sites linking here | 2 | 2 | 13 Jun 2026 to 31 Aug 2026 |
| Paid jobs traced back to a page | not measured | not measured | no click to job join exists |
The price page is the row worth copying. Search clicks fell by a quarter while lead clicks on it tripled. Fewer of the wrong people, more of the right ones.
The bottom rows are the honest ones. Click-through rate fell because impressions moved into positions four to ten, where fewer people click. At the last checks, 23 July and 31 August 2026, three directories carried a live listing: the Google Business Profile, Rehbar.pk and Ensun.io. Only the last two count as a referring domain, another site linking to yours, flat at 2 from 13 June to 31 August 2026, 11 weeks running. Nobody outside the business has linked to it in that time, which caps how far any of this can go.
We cannot tell you the desktop card worked. The desktop to mobile gap narrowed from 22.3 points to 4.0 points across three dated reads, but mobile fell too. The card logged 8 opens between 26 August and 2 September 2026, then 0 across the next 12 desktop visitors, with the tracking checked and intact.
What we would do differently, and who this is not for
Three things. We would start link building in month one: two referring domains is why a page-three ranking stays on page three. We would book the read-back dates before shipping. And we would price a chat inbox into the first scope; without one the trail stops at the conversation.
This is for you if your customers buy through a conversation and you cannot say which page started it, or if you suspect your desktop and mobile numbers are hiding each other. It is not for you if you want a ranking guarantee, or a revenue figure before a click-to-order join exists. We did not have that join here.
The cheapest change here was 139 lines of code. It came from one habit: reading conversion by device before touching a page. If your orders arrive through conversations and you cannot name the page that started them, that is the same gap.
Questions and answers
What does local SEO for a small business actually involve?
Publishing one page per thing people search for, in the words they use, then keeping the sitemap current as pages are added. Here that took the site from 11 indexable pages on 4 April 2026 to 52 on 23 September 2026.
Why do desktop visitors convert worse when the offer is a chat?
Because a wa.me link opens the WhatsApp app, and a laptop has no app. The visitor lands on WhatsApp Web and needs a paired phone. A desktop card with a scannable code, a copyable number and a web link removes that dead end.
How do you measure enquiries that finish inside a chat?
Record the click, the page and the device, then join them to booking records by hand. A button click, a unique person and a paid job are three different numbers. We report the first two and say the third is not measured.
Can you prove the search work caused the growth?
Only partly. The after window is peak dengue season in Karachi, so arrivals were always going to rise. The conversion rate is the defensible claim: a season changes who turns up, not the share of them who start a chat.
Does this transfer to a Shopify store?
Yes, where the purchase needs a conversation or a quote. The transferable parts are carrying page context into the handoff, splitting conversion by device, and booking the read-back date in advance.