Research on tenders, grants and document evidence.
Practical guides for people preparing submissions and judging document evidence.
Start with thequestion you have.
Each guide is built for search and AI answers: clear definitions, practical decision frameworks, concise FAQs, and links into the Stipple tools when the reader is ready to check something real.
Government tender search from an AI agent: the free tools, the one that costs, and coverage first
How an assistant searches open Australian and New Zealand government tenders through MCP: find_tenders and tender_sources are free with no account, match_tenders ranks them against a company’s website, and every answer states what was searched.
Read the guideGovernment tender search, and the coverage question nobody answers
Tender data is public and scattered across a dozen portals. What aggregation actually costs, why an empty search result is ambiguous, and why the source registry matters more than the count.
Read the guideHow to turn a tender pack into a requirements matrix
A tender pack is not a document, it is a list of demands scattered across several. Here is how to pull them into one matrix, with the page each line came from, before you decide to bid.
Read the guideThe tender submission checklist: what to check before you send
The checks that decide whether a finished bid is accepted: returnable schedules unaltered, addenda acknowledged, evidence current at the closing date, limits respected, lodgement mechanics right.
Read the guideHow long do you get to respond to a tender? We measured 24,424.
We measured the response window on 24,424 closed tenders. Median: 25 days. More than a third close within three weeks — and the clock starts before you find them.
Read the guideWho publishes the most government work in Australia and New Zealand
We ranked 13,032 tenders published in one year by the organisation that released them. 1,593 buyers, no dominant one, and a shire council of 14,000 people out-published a state health authority.
Read the guideWhen government work is actually released — and the two months it nearly stops
Government tenders are not released evenly. Across three independent years, August to November carries about half the year and January to April under a fifth. July is the single busiest month.
Read the guideGovernment IT tenders ask before they buy — and give you no extra time
We measured 593 government IT tenders published in twelve months. IT buyers issue requests for information at five times the rate of the wider market, prequalify at three times — then allow the same 25 days.
Read the guideWA tenders: the most concentrated, most formal market in Australia
We measured 2,428 Western Australian tenders published in twelve months. One department published 254 of them, 72 per cent are full requests for tender, and a quarter close inside 16 days.
Read the guideNSW tenders: the biggest market in Australia, and the hardest to cover
We measured 3,125 New South Wales tenders published in twelve months. 421 separate buyers, no organisation above 2.8 per cent, and a regional shire out-publishing the City of Sydney.
Read the guideSA tenders: no season, and a 27-day gap between two departments
We measured 955 South Australian tenders published in twelve months. It is the flattest market in Australia — and two departments of the same government give suppliers 15 days and 42 days respectively.
Read the guideNT tenders: a quarter of the market is getting on a list
We measured 504 Northern Territory tenders published in twelve months. Prequalification is 24.2 per cent of them against 2.0 per cent nationally — and just 64 organisations publish the entire market.
Read the guideACT tenders: the slowest and most formal market in Australia
We measured 274 Australian Capital Territory tenders published in twelve months. A 29-day median, 41 days at the upper quartile, 72 per cent full requests for tender — and almost no quick quotation work.
Read the guideQLD tenders: the fastest market in Australia, and the most local
We measured 2,274 Queensland tenders published in twelve months. Seven of the ten busiest publishers are councils, the state median is 22 days — the fastest in Australia — and one shire runs 14.
Read the guideTasmania tenders: the most evenly shared market in Australia
We measured 487 Tasmanian tenders published in twelve months. No buyer holds more than 6 per cent, councils and departments interleave, and signwriting is a bigger category than IT.
Read the guideVictoria tenders: the sharpest seasonal swing in Australia
We measured 2,010 Victorian tenders published in twelve months. July ran three times February, seven of the ten busiest publishers are councils, and grounds maintenance is a bigger category than construction.
Read the guideTender software in Australia, compared on what each one publishes
Australian Tenders publishes prices from $67 a month. TenderLink does not publish prices. EstimateOne covers construction. VendorPanel is for buyers. What each actually does, from their own pages.
Read the guideVendorPanel and Stipple: one is where councils publish, the other is where you find it
VendorPanel is a source-to-contract platform for buyers, merging with Unimarket. It is not a supplier alert service — it is one of the sources Stipple ingests. What that means if you are a supplier.
Read the guideEstimateOne and Stipple: a construction marketplace against a whole-market search
EstimateOne connects builders, subcontractors and suppliers in Australian construction. Stipple searches every public sector category. If you only bid construction, they are purpose-built and we are not.
Read the guideTenderLink and Stipple: an alert subscription against a free search
illion TenderLink claims 3,300+ sources and 41,000+ opportunities a year, and does not publish its prices. Stipple holds 2,034 tenders and is free to browse. What each is actually for.
Read the guideAustralian Tenders and Stipple: $67 a month against free browsing
Australian Tenders publishes $67, $99 and $145 monthly plans and claims 4,444 active tenders. Stipple holds 2,034 and is free to browse. The honest comparison, from their price list and our API.
Read the guideSupplier onboarding and award verification from an AI agent: the evidence gate, assembled
How an agent runs a supplier through an evidence gate before onboarding or award: company registration standing, sanctions and adverse-media screening, document verification, pack completeness and identity — each a tool call, each returning evidence, none returning a verdict.
Read the guideConnect an MCP server to ChatGPT and Claude — with a real one to try
Step-by-step: add a remote MCP server as a custom connector in ChatGPT and Claude, or by config in Claude Code and Cursor — using Stipple’s hosted server, and what works with and without an account.
Read the guideSanctions and adverse-media checks from an AI agent: what comes back, and why PEP is not screened
How an assistant screens a person or company through an MCP tool: the sanctions candidates and their strength, the adverse-media hits with evidence, the same-name exclusions, and why PEP is reported as not screened.
Read the guideHow adverse-media and sanctions screening data is collected — the sources, the gaps, and the judgement
Where a screening result actually comes from: four primary sanctions lists taken from the issuers with their own dates, why the UN list and PEP data are absent, how the news layer finds and judges an article, and what a weekly rebuild promises.
Read the guideHow document fraud detection actually works — 43 checks, and why 17 of them are not about fraud
What a forensic document check actually inspects: 43 signals across two axes, why arithmetic beats appearance, and why a clean result is coverage rather than a certificate.
Read the guideWhy AI agents get tables wrong — and what to do instead
Agents misread tables because they are shown a picture of one. What rasterisation throws away, why a native PDF already holds the answer, and how to test any extraction tool.
Read the guideSanctions and adverse-media screening, for developers who have to justify the result
How a screening API decides a name matches: corroboration gating, why a name-only hit is never strong, what same-name exclusion looks like, and what this data does not cover.
Read the guideWe removed a watermark from a PDF. Here is what it left behind.
We watermarked a PDF, removed the watermark three different ways, and ran every version through document forensics. The removal trail — revisions, redefined objects, modification metadata — and the one method that leaves none.
Read the guideOne in 771: AI-generated evidence has reached AFCA — barely
We searched the full text of 771 published AFCA determinations (June–August 2026) for generative-AI phrases. Exactly one mentions AI-generated documents — and it shows how an ombudsman spotted them, and why they were refused.
Read the guideAI fact checker: how to verify an AI-written report, step by step
What an AI fact checker can actually verify — citations, arithmetic, contradictions — and what it cannot. A step-by-step workflow for checking AI reports, the best fact-checker websites by job, and the measured rate of fake citations.
Read the guideClaude’s invisible text watermark: what it proves, and what it doesn’t
Anthropic is embedding invisible watermarks in Claude-generated text from August 2026, using a version of SynthID-Text. What the mark proves, what removing it takes, and why absence proves nothing.
Read the guide39 of 101 citations in AI-written reports did not exist
We asked language models for referenced reports and checked every citation. 38.6% did not exist as cited — and the rate doubled when the topic thinned out.
Read the guide94.7% detection across 12 language models
We generated documents from 12 language models and detected 94.7%, then measured what most vendors leave out: false accusations of human writers.
Read the guideFour payslip forgery types caught every time, one we gave up
We forged 192 of 210 Australian payslips six ways and ran each through the same forensic checks. Four types were caught every time. One we gave up.
Read the guideThe 10 best AI detectors in 2026, compared honestly
Ten AI detectors compared on what actually matters — explainability, false positives, abstention, access — with an honest methodology, a comparison table, and the failure modes every vendor shares.
Read the guideAI detector guide: what the score means and what to do next
A plain-English guide to AI detectors, AI checker tools, free AI detection, false positives, and how to review AI-writing results fairly.
Read the guideAI hallucinations: why models make things up — and the one kind you can check
What an AI hallucination is, why models invent facts and sources, real examples — and the one hallucination you can verify: citations that don’t exist.
Read the guideHow to humanize AI writing so it actually sounds like you
A beginner-friendly guide to humanizing AI text, improving AI writing, using AI humanizer tools responsibly, and avoiding generic AI-sounding prose.
Read the guideAI paraphrasing guide: rewrite clearly without changing the facts
A clear guide to AI paraphrasing tools, paraphrasing without plagiarism, preserving meaning, and checking rewritten text for accuracy.
Read the guideAI image detection: how to check if an image might be AI-generated
A simple guide to AI image detectors, image provenance, C2PA, metadata, synthetic image signals, and how to check an AI-generated image responsibly.
Read the guideDeepfake detection: how to spot synthetic media without overreacting
A practical guide to deepfake detection, synthetic media, provenance, and how to verify suspicious images, documents, video, or audio claims.
Read the guide