Data

Most sites publish a number and hide the method. Three measurements here publish the method first: what gets counted, over what sample, how often. Below them is every market figure this site prints, each with the date it was checked at source and a mark saying how far to trust it.

The datasets

Each of these is produced by work already happening rather than by work commissioned to fill a page, which is the only reason one person can sustain a research surface at all. The method is published first so the number cannot be quietly shaped to fit it.

01

What AI answer engines cite for AI app builder questions

When somebody asks ChatGPT, Perplexity, Claude or Gemini which AI app builder to use, which sources does the answer come from?

No finding yet. Nothing has been measured.

Method
A fixed set of 20 queries run weekly across four answer engines. Every cited source is logged with its URL, its publisher and the date of the run. Nothing is aggregated across engines: they retrieve differently and averaging them would hide the only interesting variable.
Sample
20 queries, 4 engines, weekly
Cadence
Run weekly, published monthly
Blocked by
Nothing. The loop is already running as an internal check, and publishing it costs one export.
02

The real ₹ cost of shipping a student project with AI

What does it actually cost, in rupees, to take one small project from nothing to a live URL that stays up?

No finding yet. Nothing has been measured.

Method
Measured off published build pages rather than estimated. Every build records what was spent to build it and what it costs per month to keep running, including the free-tier limits that keep the second figure at zero and the point at which each tier stops.
Sample
Every published build, no exclusions
Cadence
Recomputed whenever a build ships or a provider changes a tier
Blocked by
Builds. The figure is meaningless below roughly ten of them, and there are none today.
03

Where AI coding tools fail, categorised

When an AI builder gets something wrong, what kind of wrong is it, and how often?

No finding yet. Nothing has been measured.

Method
Aggregated from the 'Where it broke' section of every build page and the 'The Catch' section of every case study, sorted into a failure taxonomy with counts. Every entry keeps a link to the write-up it came from, so any count can be traced back to the specific failure that produced it.
Sample
Every documented failure across builds and case studies
Cadence
Recomputed on each publish
Blocked by
The same thing: no builds and no case studies means no failures to categorise.

Figures this site prints

These are other people's numbers, not our measurements, and the distinction is the point of keeping them on a separate list. Each carries the date it was checked at source rather than the date it was published. Two figures were retracted from our own documents on 2026-07-30 for being unverifiable, and neither appears anywhere on this site.

41-46%

of new code written globally is generated by AI.

checked 30 Jul 2026verified at source

4:1

People who build business applications without being developers, against professional developers. Roughly 100-120 million to 27.7 million.

checked 30 Jul 2026verified at source

70%

of new enterprise applications will use low-code or no-code by 2026, up from under 25% in 2020.

Gartnerchecked 30 Jul 2026verified at source

750+

no-code platforms exist. Almost none of them will tell you which one to start with.

checked 30 Jul 2026verified at source

63-84%

of AI app-builder users are said to have no coding background.

checked 30 Jul 2026no method published

Two published figures, neither of which states a method. Directionally consistent, individually unreliable. It is the most useful number on this page and it is the one you should trust least.

Prior work

Three published studies this site's method rests on, linked rather than paraphrased. Each row states what the study found, how large its sample was, when we checked it at source, and what this site does differently because of it.

  1. Do AI assistants prefer to cite fresh content?

    The average cited URL is 1,064 days old across AI surfaces against 1,432 days in organic Google results, making AI citations 25.7% fresher. ChatGPT is freshest at 958 days. Roughly half of AI citations point at content under 13 weeks old.

    Ahrefs16,975,000 cited URLs across six surfaceschecked 30 Jul 2026

    Every page here carries a visible updated date, and that date is bumped by hand on substantive change rather than by the build. The recency window that wins citations is months, not years.

  2. We analysed 137,000 sites to see if llms.txt does anything

    97% of llms.txt files received zero traffic in May 2026. Across 500 million AI bot visits over 90 days, 408 targeted llms.txt.

    Ahrefs137,000 siteschecked 30 Jul 2026

    This site keeps its llms.txt because coding tools genuinely read it, and spends no maintenance budget on it. Effort that would have gone there goes to off-site corroboration instead.

  3. Build and submit a sitemap

    Google ignores the priority and changeFrequency attributes entirely. lastmod is the only sitemap attribute it acts on, and only while it stays verifiably accurate.

    Google Search Centralchecked 30 Jul 2026

    Our sitemap declares neither. Mechanical date-bumping to fake freshness is the fastest way to make the one signal we control worthless.

How to cite this

Attribution that takes effort does not happen. This is the line, ready to copy.

ALGOBIC. "Data". https://www.algobic.in/data. Retrieved 2026-08-09.

Quote any figure on this page with its confidence mark attached. The mark is part of the figure. A number of ours reproduced without the sentence saying nobody published a method for it is a number we did not print.

Where next

AboutWhy now, in the same figures, with the same dates attached.Case studiesWhere the failure transcripts come from that dataset three counts.BuildsWhere the ₹ figures come from that dataset two measures.

Think one of these numbers is wrong?

Say so. Two figures have already been pulled from our own documents for being unverifiable, and every correction gets made in public with the date on it. Instagram is the fastest way to reach a person here.

@algobic.in