Public review log

What we changed, and why

A directory of other people's job pages is only worth keeping if somebody keeps checking it. This page is the record: every structural change, every page we deleted, every rejection we received and what we did about it. No date is back-dated and no entry is written as if it went better than it did.

Maintained by Gopal Chandrawanshi · first entry 15 Jun 2026 · this page updated 20 Oct 2026

Listings tracked

all linking to the employer's own page

Last rate sweep

28 Sep 2026

next full re-verification: 30 Nov 2026

Pages retired after review

139

93 interstitials (Oct 9) + 46 near-duplicates (Oct 18), each with a redirect or a real 404

Corrections published

6 since Jun 2026

each answered by fixing or deleting the page, never by adding a new one

How this directory is maintained

Verification, not recycling. A listing is only published after we open the company's own recruiting page. We do not copy listings from other job boards, blogs or Facebook groups, and we do not write rates that the company has not published somewhere itself.
Quarterly sweep. Every quarter we re-open all 46 pages: rates, open-role counts, whether the platform is still taking contributors at all. Anything we cannot reach is removed rather than left up with a stale date — the entries below are the audit trail of that policy working.
Every listing carries its capture date on the card. When the date is more than two months old, treat the figure as historic and check the platform's page; we would rather say "this is when we saw it" than pretend it is live.
Corrections are honoured and logged. If a platform changes a rate, closes a programme, or our summary is wrong, email [email protected]. We fix it and record it here, including who reported it (with permission).
Deletion over padding. When a page stops earning its place — near-duplicate, thin, or a redirect wrapper — we delete it and return a real 404. This log exists partly because we would rather document 139 deletions than hide them.
What we will not do: generate "news" around job listings to look active, auto-post summaries of third-party listings, or keep an opportunity listed because a referral link pays us. Listing decisions and commercial relationships are separated, and the editorial rules are in the editorial policy.

Log

One page, one version of each sentence — 2026-10-20

After the 18 October card rebuild, the homepage carried 24,600 words of visible text but only 387 unique sentences: the four-stage process, the Day 0 → Week 4–8 timeline and the safety note were printed in all 46 cards, so 10,500 words were literal duplicates. That is the pattern Google's spam systems and AdSense reviewers describe as low-value, repeated content, and it was also the largest object on the page (384 KB HTML).

The duplicated blocks are now one shared "How hiring actually works" section above the guides, with the pay-reality and safety notes stated once. Every card keeps what is specific to it: the company summary, who it suits, what it requires, what to check, what to keep in mind, its sources and its apply link. The page is 289 KB and 11,900 words, 99% of which is unique. Nothing was deleted from the record — the shared text simply appears once.

Same day: old .html URLs now 301 to the clean URL instead of serving a second copy of every page, the 46 deleted /jobs/*.html addresses 301 to their card anchors instead of returning 404, the homepage counters no longer start at zero in the HTML, and the sitemap lastmod values reflect the files that actually changed.

Card breakdown restored — content first, apply last

2026-10-18 · same day, after the deleted job pages shipped

Deleted pages used to hold ~900 words per platform, and the apply link sat at the bottom of them. The cards kept only the summary, so the reading order inverted: one click and you were off-site. Every card now carries the full funnel inside it — what it is, who it suits, what to check before applying, the four-stage process, a realistic Day 0 → Week 4–8 timeline, the honesty note, what to do before handing over documents, then related guides, and the apply link last.
Two process variants instead of one shared template: project/task-platform stages vs employer hiring-loop stages, because those are genuinely different sequences.
Fixed a desktop clipping bug found while doing this: .jcard.open .jdetails was capped at max-height:800px with overflow:hidden, so long breakdowns were cut off mid-sentence on screens over 640px. The cap is gone (fade instead of a clipped slide).
Cards are now addressable: howtoaijobs.com/#job-scale-com-… jumps to the right page of the list and opens that card. The 46 deleted /jobs/<platform>.html URLs 301 to those anchors instead of 404ing, so existing inbound links and old search results still land somewhere useful.
Share and copy-link were inheriting the employer URL from the apply button — i.e. a share said "found on HowToAIjob" and then sent people to terac.com/login. They now share our own card anchor.
Header affordance: cards were labelled clickable but the cursor was suppressed on desktop with no other cue, so they read as inert. Added a visible "Full breakdown" toggle label, a hover tint and a focus ring.
Removed two stale third-party verification metas (Pinterest and a Facebook domain verification for a file that is no longer used) from the homepage head.

Structural cut: 46 near-duplicate job pages removed; benchmark data + two tools published

2026-10-18

Deleted every /jobs/<platform> page (46 files). Each one carried roughly 116 words of its own text and 1,400 words of shared boilerplate — 93% identical to its neighbours. The same information was already on the homepage card for that platform, so nothing is lost by deleting them; a second copy of everything, spread across 46 URLs, is exactly what "mass-produced content" means.
Replaced the JobPosting structured data on the homepage: 44 of its 46 url fields pointed at pages that no longer existed. It is now an ItemList pointing at each employer's own page, because a JobPosting is only valid on a page that hosts the description and the application — which this directory deliberately never does.
Every "View details" button became an explicit "Apply on <platform>" link with rel="nofollow noopener noreferrer sponsored". There is no interstitial and no redirect, and the link now says what it is.
Removed fake freshness. Cards said "Today", "1 week ago", "1 month ago" from a static file — the listing date is now shown as the date we actually captured it, matching the verification note on each card.
Fixed the hero counter printing 100%% (a "%" in the JS plus a "%" in the label) and the label "% Remote-friendly".
Rewrote 12 job titles that were company slogans rather than roles ("Powers 90% of World's Top AI Models", "world's #1 computer vision tool", "$10k-$1M/yr top 1% remote jobs"). If we print that language, we own it.
Published the AI Work Rate Index — all 46 platforms, one sortable table, with the capture method, the 2,080-hour conversion for annual quotes, and every "not published" left as a hole.
Published the AI Work Pay Checker — a browser-side estimator that compares a claimed rate against that index and applies payout fees and bench weeks.
Started this review log. It is the only honest answer to "is anyone maintaining this?".
Expanded the two shortest indexed guides (Philippines and Nigeria) and cleaned stale rules out of robots.txt; sitemap.xml now lists only pages a person would want to land on.

"Low value content" round 1 — doorway purge (PR #6, #7)

2026-10-09

Deleted 93 orphan interstitial pages (~134 words each, "You are leaving HowToAIjob" templates) that were crawlable, ad-enabled and linked from nowhere.
Rewrote the hire-talent page from 201 words of claims to ~900 words about how the listing process works, what a good listing contains and what we cannot promise.
Expanded 3 thin guides past 1,000 words each with pay tables, onboarding reality and expectation warnings.
Removed keyword stuffing from the homepage paragraph and meta description.
Trimmed sitemap.xml from 63 URLs to 18 so only substantive pages are submitted.

Rejections on howtoaijobs.com — first content clean-up

2026-08-13 → 2026-07-19

Removed the auto-redirect + countdown from go.html (click-through interstitials are prohibited).
Replaced the fake auto-generated "today" date with a real, static review date.
Counters became real DOM counts instead of hardcoded 1,000 / 100%.
Rewrote 49 hype claims; linked every job card to its own page; added the Guides section (10 articles) and an affiliate disclosure.
Added JobPosting + Article + BreadcrumbList schema, OG tags, real 404 page, and a _redirects rule so unknown URLs return 404 instead of the homepage.

Rejections on the pages.dev preview domain

2026-07-10 → 2026-08-01

Application first submitted for howtoaijob.pages.dev; rejected twice. Mapped the root domain howtoaijobs.com and resubmitted there, because a preview subdomain on a free host is not a site Google can assess as a durable publisher.

Site created, AdSense account opened, application submitted

2026-06-15 → 2026-06-18

Static directory launched on Cloudflare Pages; ads.txt published with our publisher ID; privacy, terms, disclaimer, editorial policy and contact pages written before the application, not after it.

Open items we have not finished

Re-verify all 46 listings against live recruiting pages (target 30 Nov 2026). Every card currently carries the 28 Sep 2026 capture date, which is honest but now ageing.
Rate figures were captured 28 Sep 2026 and are labelled as such on every card; the next sweep moves the date and updates the index table in one pass. Until then the honest statement is "captured 28 Sep 2026", not "live".
Per-row source URLs in the Rate Index. The method page explains how rates were captured; the next revision will link each captured figure to the specific page it came from, so a reader can check one row without re-tracing the whole sweep.
Country guides are only written where we can state facts about payment rails and eligibility. Three more are drafted but unpublishable at that standard, so they are staying unpublished.
Evidence of real demand. Traffic is not something we can write into a page — we are adding the things people actually use (the index, the checker, corrections by email) and letting the log show the maintenance.

Advertising appears on this site. This log describes the site's own structure and editorial decisions; it is not a marketing page and it is not updated retroactively without an entry.