SME AI Hiring Index · Methodology · updated October 2026

How the Index works

The Index is a monthly sample of AI and automation job ads visible to Melbourne job seekers. It records what sampled ads say. It is not a census of the labour market or a measure of AI adoption across all businesses.

How ads are collected

  • Ads are captured manually from SEEK, LinkedIn and Indeed using searches for AI, automation and the operational technology titles that often contain this work.
  • The June 2026 baseline contains 45 ads captured on 30 June. The July edition contains 52 captures from 17–27 July, including three roles seen in June. The August edition contains 46 captures from 11–30 August, representing 45 distinct roles after one cross-board duplicate was removed. The September edition contains 40 captures from 3, 25 and 28 September, representing 39 distinct roles after one re-advertised July role was identified.
  • The location filter is Melbourne. Some ads are remote or list Melbourne alongside other cities, so the Index describes ads visible to a Melbourne job seeker rather than every role being based only in Melbourne.
  • Search terms can change as employers change how they name the work. Editions are therefore comparable observations, not identical samples or a basis for month-to-month trend claims.
  • Each captured ad is preserved and converted into a structured record before the monthly analysis.

How ads are classified

  • Employer size comes from public platform bands first, stated headcounts second and structural evidence third. When the ad and the board give none of these, we look the employer up after capture; most smaller employers are sized this way. Businesses under 200 staff are SME-flagged. Unverifiable employers stay in a separate bucket.
  • Employer type separates direct employers, consultancies, AI vendors and recruiters. This matters because a vendor hiring AI staff is evidence of supply, not necessarily its own AI adoption.
  • What the role does is taken from the ad's duties and evidence, not the job title alone. An AI term in a title, tool list or candidate requirement does not by itself establish an AI programme.
  • Programme stage and audience are counted only when the ad describes the programme clearly enough to classify. Supply-side roles and ads with no described programme are excluded from those figures. From September 2026, stage is reported only as which stages appear. Counts by stage haven't stayed the same when ads were re-read, because where an ad sits between two stages is a judgement call.
  • Salaries and tools are counted only when the ad states them. Every published figure includes its sample size or disclosed count.

What the Index cannot tell you

  • The sample is manual, Melbourne-focused and shaped by the search terms. It is not a random or complete sample of the Australian labour market.
  • Job ads show what employers wrote and intended. They do not prove that a system has shipped, works reliably or produced the stated outcome.
  • Only employers advertising appear. Businesses using contractors, consultancies or existing staff may not appear at all.
  • Employer size is inferred from public evidence and can be wrong in individual cases. Uncertain cases are excluded from size-based findings.
  • Small subsamples are reported as counts and worked examples, not population rates. Three editions can show a recurring pattern in these samples, but not a market-wide trend.

Corrections and updates

Pages carry dated notes on what changed and why. This list collects them, newest first. The June and July pages list their own changes in more detail.

  • 15 October 2026.
    • The June and July pages said we read “every AI and automation job ad we can find”. Both are corrected to “a manual sample”.
    • Every page, and every edition's share image, now describes the frame the same way: Melbourne job ads, captured with a Melbourne location filter. “Melbourne-weighted” and “Melbourne-area” are retired.
    • The June page and its search description said none of June's four small businesses “advertised an AI job title”. That was wrong when it was added on 28 July, because two of the four titles include “AI”. It's corrected to “none advertised an AI-first title”, and the search description no longer makes the claim. The June page now also lists its 28 July headline changes, which weren't disclosed before.
    • The small-business job-title observation (June, July and the SME-vs-enterprise analysis) hasn't been consistent: all five August ads from employers under 200 staff had “AI” in the title; four of September's six didn't.
    • The July and August stage tables carry a note: read them only as showing that all four stages appear. The June page notes its restated stage figures. The July page text and the landing page no longer quote a planning-or-piloting count.
    • The July page now lists each version of its headline and figures since first publication.
    • August's audience split is restated from 23 internal only and 8 internal and customer-facing to 22 and 9. One programme was recoded under the written audience rule; the headline count, 31 of 32 involving internal work, doesn't change. The landing page carries a matching note. A note on the August page also marks its supply-side comparison as provisional.
    • The SME-vs-enterprise search answers are reworded. They no longer say direct hiring is “nascent” or name consultancies as where SME demand lands, and June-only observations are labelled as June's. One body sentence about consultancies is softened to “may be”.
    • Publication dates are corrected to the day each page was first committed, in Melbourne time: July 3 August (was 2 August) and August 3 September (was 1 September). June, the SME-vs-enterprise analysis and this page keep 4 July.
    • The June, July and August share images are replaced, so they show each page's current headline, counts and frame. June's showed a retired “Only 9%” headline. They're published under new file names, so cached copies of the old cards aren't reused, and the old SVG versions are removed.
    • The June page now lists its 4–5 July changes. It was first published on 4 July as the “July 2026 Baseline Report”, headlined “Only 4 of 45”, and was relabelled June and moved to its current address the same day. The old address now redirects to it.
    • August's headline “None point at customers alone” now reads “None are customer-facing only”, as does the landing page's archive line.
    • July's methodology set-aside count is corrected: six ads with no AI programme and three repeated June roles, not eight and one. The 32 programmes don't change.
    • Two June sentences are reworded: “employers at the front of the queue” became “employers who are advertising”, and the fractional-leadership advice no longer calls it a “permanent” capability.
  • 4 September 2026. Every June and July ad was re-read under the August classification rules.
    • June's audience figure went from 29 of 29 to 30 of 30 programmes involving internal work.
    • July's split is now 20 internal only, 8 internal and customer-facing, 3 customer-facing only and 1 unclear, with 28 of 31 still involving internal work.
    • Stage rows were restated. June went from 9 / 8 / 4 / 8 to 9 / 5 / 7 / 12, and July from 10 / 7 / 7 / 7 to 13 / 5 / 11 / 3.
    • On the August page, published the day before:
      • “three more June ads describing a programme” became “four more June ads describing a programme, three of which were too vague to classify by audience”;
      • the headline's “AI programmes described in August's job ads” became “AI programmes in August's employer-adoption sample”;
      • the reason for setting 13 ads aside changed from “because the employer sells AI” to “12 from consultancies or AI product businesses, and one role that combined internal and client delivery”.
  • 5 August 2026.
    • July's programme count was corrected from 32 to 31, and its customer-facing-only count from four to three, with other figures on the July page. These were later superseded by the 4 September re-read.
    • June's capture date was corrected from 3 July to 30 June 2026 on the June, SME-vs-enterprise and methodology pages, and June's “fixed sample” became “manual sample”.
    • This page dropped the limitation “AI-title keyword collection undercounts SMEs, whose AI duties often sit inside operational titles”.
  • 4 August 2026. July's headline was reworded. Several July figures changed in the same edit; the July page lists them.
  • 28 July 2026. The June page's headline changed twice, percentages became counts, and the job-title claim later found to be wrong was added. The June page lists these changes.
  • 4–5 July 2026.
    • The baseline edition, first published on 4 July as the “July 2026 Baseline Report”, was relabelled June and moved to its current address. Its headline moved from “Only 4 of 45” to “Only 9%”.
    • The June and SME-vs-enterprise pages were rewritten on 5 July to treat small businesses as peers of large ones. The SME-vs-enterprise opening line calling small-business AI awareness “nascent” was removed, its headline changed, and seniority advice was added to both pages.

Cite this index

GraftPoint SME AI Hiring Index, September 2026 edition. Manual sample of 40 Melbourne AI and automation job-ad captures (Melbourne location filter) from 3–28 September 2026, representing 39 distinct roles. Methodology: graftpoint.com.au/sme-ai-hiring-index/methodology/

Questions about the method, or want the underlying counts for a story? Email june@graftpoint.com.au.