The problem: loose pages, no hierarchy
The keyword map was built with AI, with no paid tools and no SEO criteria. The site ended up with no hierarchy: loose pages, each on its own, with no topic grouping them.
The diagnosis: the content answered searches almost nobody made
The keywords came from conversations with the client and from AI tools, and that map was handed to Base44 to build the site. It's the first keyword research (spreadsheet, in Spanish): search intent and a suggested page, but no volume.
There was a rush to rank, and the content came out informative but dry. To Google and to a reader, the pages looked like a collage, and a busy design didn't help reading them either.
The fix: research with data and organize the site by topic
With DinoRank, DataForSEO, Semrush, Cloudflare and Claude:
Keywords with volume and difficulty: second keyword research (spreadsheet).
Competitors: their traffic, the formats that work for them and the tools that bring them visits. Competitor analysis.
Topic groups: each with its pillar page (the general page on the topic, which links to the specific ones). Whatever changed address was redirected with a 301 so nothing already earned was lost.
Migrations with Cloudflare and Claude Code: the 301 redirects that made the new architecture possible live in a Cloudflare Worker. Claude Code generated its code; all 12 returned a single-hop 301 (curl, Sep 24, 2026).
Content that answers up front: each page answers the search in its first lines. Content production.
A tool of our own: a calculator for late-payment interest owed to the Colombian tax authority (DIAN), specified by me and generated with Claude Code. People were already searching for it, and competitors got traffic from tools like it. “liquidador intereses de mora dian 2026” alone gets 4,400 searches a month in Colombia (Google Ads via DataForSEO, October 2026).
Every search has a single page that answers it. This is how it looks in the research:
12 of the 96 searches, in 9 groups, from the second keyword research (the project's working sheet, read on Oct 31, 2026). Volume and difficulty for each are in the sheet. Searches are in Spanish, as people type them.
| Search | Intent | Group | Page that answers it | Conversion potential |
|---|---|---|---|---|
| liquidador intereses de mora dian 2026 | Transactional | Tools | /herramientas/calculadora-intereses-mora-dian | Medium |
| ley 1116 de 2006 | Informational | Law 1116 · grounds and legal framework | /ley-1116 | Medium |
| ley de insolvencia persona natural | Informational | Personal insolvency | /insolvencia-persona-natural | High |
| ley 2437 de 2024 | Informational | Law 1116 · grounds and legal framework | /ley-1116/ley-2437-de-2024 | Medium |
| prueba ácida | Informational | Tools | /herramientas/autodiagnostico-insolvencia | Medium |
| desventajas de la ley de insolvencia | Commercial | Law 1116 · the process and decisions | /ley-1116/desventajas-mitos | Medium |
| proceso de insolvencia | Informational | Law 1116 · the process and decisions | /ley-1116/guia-del-proceso | Medium |
| reorganización empresarial | Commercial | Main pages | / | High |
| cuanto cuesta un proceso de insolvencia en colombia | Informational | Personal insolvency · cost of the process | /insolvencia-persona-natural/cuanto-cuesta | High |
| cobro juridico | Informational | Personal insolvency · debt collection | /insolvencia-persona-natural/cobro-juridico | High |
| embargo cuenta bancaria | Informational | Warning signs and emergencies | /blog/embargo-cuenta-empresarial-dian | High |
| asesoría ley 1116 bogotá | Transactional | Main pages | /contacto | High |
The calculator exists because of the first row: someone searching for a calculator wants to run the numbers, not read an article. And the conversion column decides which pages carry the call to book a first call.
How the architecture changed
Before · version 1
Home page
Hanging from the home page8 pages
Law 1116 · Law 1116 and credit history · Reorganization vs. liquidation · Business reorganization guide · Emergency route · About Carlos · Contact · Blog
Hanging from the blog8 pages
Payment default · Imminent inability to pay · Law 2437 of 2024 · Insolvency self-assessment · Signs of financial crisis · Unpaid employees · Unpaid suppliers · Tax authority account seizure
The Law 1116 topics were split between the home page and the blog, with no page grouping them.
After · Oct 31, 2026
Home page
Law 1116 (pillar)10 pages
How to avoid the process · Reorganization vs. liquidation · Imminent inability to pay · Drawbacks and myths · Law 2437 of 2024 · Emergency route · Payment default · Credit history · Process guide · Winding up a company
Tools3 pages
Tax interest calculator · Insolvency self-assessment · Projected cash flow
Personal insolvency (pillar)5 pages
Process · Asset liquidation · Law 2445 of 2025 · How much it costs · Debt collection
Blog5 pages
Signs of financial crisis · Tax authority account seizure · Unpaid suppliers · Unpaid employees · Tax authority payment agreement
About Carlos, Contact and the legal pages sit outside the groups.
Drawn from the Octopus.do maps exported on Oct 15, 2026 and from the sitemap served on Oct 31, 2026 (30 URLs; the personal insolvency group went from three pages to 5).
Before: almost everything hung from the home page or the blog
Octopus.do · the architecture of version 1, exported on Oct 15, 2026 and captured in full on Oct 31, 2026.
One Law 1116 page could sit under the home page and the next one under the blog. Neither Google nor the reader could tell which one was the general page.
After: groups with their pillar page
Octopus.do · the current architecture, zoomed in, screenshot of Oct 31, 2026. Each box carries the page's title and description.
Each group has its general page on top and one specific question per page underneath.
See the detail of the Law 1116 group
The redesign, as part of the SEO. Before: a dark background, a dim photo and many calls to action competing. After: a light background, a headline that says what's offered and a single photo.
Less noise and larger text: the content and the architecture show when reading, on a phone too.
The mobile design decisions, along with the landing page that filters clients, are in the web design and funnel case study.
Grade A on security headers
Security Headers · Oct 14, 2026: five of the six security headers, active.
It shows the server was configured too, not just the content. The missing one, the CSP, is in testing before it's switched on (technical review).
I added the yellow box.
The result: 11 pages between positions 5 and 9, and a tool AI cites
Average position in Google by page
Imminent inability to pay
5.1
63 impressions · 1 clicks
Personal insolvency process
6.1
64 impressions · 2 clicks
Law 2437 of 2024
6.7
306 impressions · 3 clicks
Payment default
7.0
142 impressions · 1 clicks
Self-assessment (acid test)
7.3
569 impressions · 0 clicks
Credit history
7.7
123 impressions · 4 clicks
How much personal insolvency costs
8.1
145 impressions · 7 clicks
Tax authority account seizure
8.1
209 impressions · 2 clicks
Unpaid employees
8.7
112 impressions · 2 clicks
Personal insolvency
8.9
425 impressions · 3 clicks
Unpaid suppliers
9.4
103 impressions · 1 clicks
Pages with 50 impressions or more and an average position below 10. On the scale from 1 to 10, further left means higher in Google. Search Console, Oct 2 to 29, 2026, read on Oct 31, 2026. On Oct 15, 2026, with the Sep 16 to Oct 13 window, there were 9.
6 → 30 clicks · 336 → 2,805 impressions
Search Console · Jul 12 to Aug 12 vs. Sep 14 to Oct 14, 2026.
Still few clicks, and I say so. What changed is that the site now shows up where people search for what it offers.
I added the yellow box.
In those weeks three things changed at once: the keywords inside the content, the answer at the start of each page and a useful tool. The result can't be credited to any single one.
The calculator, cited by Bing's AI
People weren't looking for an article: they were looking for a calculator. Building the tool the search asked for got Bing and Copilot citing it about 190 times a day since Sep 28, 2026. That's 6,087 citations in 32 days, vs. 75 in the previous 34 days.
Citations per week in Bing's AI answers
Full weeks, Monday to Sunday. Almost no citations until the week of Sep 21; 704 the week of Sep 28 and over a thousand every week of October.
Source: Bing Webmaster Tools, AI Performance, daily export from Aug 25 to Oct 29, 2026, read on Oct 31, 2026. The week of Oct 26 is incomplete and isn't drawn.
Show the chart values
| Week of | Citations |
|---|---|
| 2026-08-31 | 18 |
| 2026-09-07 | 17 |
| 2026-09-14 | 5 |
| 2026-09-21 | 35 |
| 2026-09-28 | 704 |
| 2026-10-05 | 1675 |
| 2026-10-12 | 1224 |
| 2026-10-19 | 1543 |
Citations come Monday to Friday: in October, 222 to 313 a day on average, vs. 55 on Saturdays and 35 on Sundays. People use it while they work.
Of the listed citations, 96.5% come from calculation searches: “liquidador de intereses moratorios dian” alone accounts for 3,850 (72%).
With almost the same impressions, Bing brought more clicks than Google. From Aug 25 to Oct 29, 2026, Bing search gave 82 clicks and 6,498 impressions (Bing Webmaster Tools). Google gave 63 clicks and 6,569 impressions (Search Console).
In Bing's keyword report (web search only, 3 months, exported on Oct 31, 2026), 64 of the 66 clicks come from calculator searches. And the search with the year performs better: “liquidador de intereses moratorios dian 2026” had a 7.1% CTR, and without the year, 0.4%.
A citation isn't a visit, and a visit isn't a client: these figures don't count clients.
On Google, AI shows the guides
Google is a different story. Its AI features (AI Overviews and AI Mode) mostly show the law guides: the Law 2437 of 2024 guide alone has 209 of 758 impressions, and the calculator, 27. Two search engines, two kinds of answers: that's why the architecture has both, guides that explain and tools that calculate.
Those impressions went from 79 by Sep 17 to 758 by Oct 28, 2026. This report has no clicks, so it doesn't say how many visits they brought.
Impressions per week in Google's AI features
Full weeks, Monday to Sunday, from Aug 24 to Oct 25, 2026. From 10 to 29 a week until mid-September; from 76 to 139 since the week of Sep 21.
Source: Search Console, Performance › Generative AI features, prefix property, export from Aug 22 to Oct 28, 2026, read on Oct 31, 2026. The last three days, Oct 26 to 28, add up to 145 and don't make a full week.
Show the chart values
| Week of | Impressions |
|---|---|
| 2026-08-24 | 10 |
| 2026-08-31 | 20 |
| 2026-09-07 | 29 |
| 2026-09-14 | 29 |
| 2026-09-21 | 85 |
| 2026-09-28 | 137 |
| 2026-10-05 | 139 |
| 2026-10-12 | 76 |
| 2026-10-19 | 87 |
758 impressions in Google's AI features
Search Console · Performance › Generative AI features, Aug 22 to Oct 28, 2026, screenshot of Oct 31, 2026. The report only exists in the interface and has no clicks.
Colombia accounts for 638 of the 758, and desktop for 516: most people search from a computer.
I added the yellow box.
For an AI to cite a page, it first has to be able to read it. In the 30 days to Oct 31, 2026, crawlers requested the site about 6K times, and none of them is blocked:
Googlebot 3.42K · ClaudeBot 440 · BingBot 320 · Applebot 210 · OAI-SearchBot 180
Cloudflare AI Crawl Control · successful requests by crawler, last 30 days, screenshot of Oct 31, 2026.
Googlebot requests the most, but ClaudeBot, BingBot, Applebot and OpenAI's search crawler read the site too. That's why every page answers in the HTML, without waiting for JavaScript.
I added the yellow box.
No crawler blocked
Cloudflare AI Crawl Control · per-crawler policy, screenshot of Oct 31, 2026. The figures in each row cover the last 7 days.
Blocking AI crawlers takes the site out of their answers. The decision here was to let all of them in.
I added the yellow box.
Next steps: more channels, and measuring who is a client
The site is still young: to grow faster it needs channels beyond SEO. And not everyone who uses the calculator is an insolvency client: how many are still needs measuring.
The calculator's next improvement: a block at the top with this month's rate, because people search for the current year's rate.