Blank evaluation template. No AI trial results or employer feedback are claimed here. This is Luke's review record, separate from the employer experience.
Controlled run
Create one pair of runs per vacancy/platform/model/tier combination. Use the same vacancy facts and selected outputs. Start each run in a fresh chat, use the same available AI settings, and follow each method's intended instructions. Do not simplify only one method or silently add missing facts to it. Record any unavoidable difference in tools, memory, browsing, attachment support or skill setup.
| Run details | Record |
|---|---|
| Run ID, date and tester | |
| AI assistant, model/version, subscription tier | |
| Settings/tools available; fresh-chat confirmation | |
| Vacancy scenario and source facts | |
| Selected outputs | SEEK / LinkedIn Jobs / LinkedIn feed / Facebook organic / short notice |
| Baseline file/version | Supplied v1.4 skill; exact filename/hash if available |
| Formula version/commit | |
| Order used and any deviations |
Save the completed brief, relevant baseline inputs, both transcripts and both unedited first outputs. Preserve later revisions separately. Use fictional or approved example vacancy data.
Measures
Start timing before the method's required preparation, including reading/uploading/installing instructions or completing the builder. Stop at the first reviewable output, then separately record the time and effort to obtain an acceptable final result. Exclude interruptions consistently and record them. Count preparation only once where it is truly reusable, while reporting that initial cost separately from repeat-vacancy cost.
| Measure | Supplied v1.4 skill | Formula | Notes/evidence |
|---|---|---|---|
| Initial setup time | |||
| Vacancy entry + first output time | |||
| Total time to acceptable result | |||
| Human actions: clicks/selections/uploads/copy-pastes | Count using the same convention | ||
| User messages / AI turns | |||
| Human-authored input words | Exclude bundled system/skill/export instructions; record pasted words separately | ||
| Re-entered facts / repeated questions | |||
| Revision requests after first output | |||
| Selected outputs produced / requested | |||
| Field text blocks present / expected | Score against the chosen format, not generic ad quality | ||
| Manual platform selections clearly separated | |||
| Facts correct / incorrect / invented | List each material discrepancy | ||
| Missing facts and conflicts correctly flagged | |||
| Required safeguards missing | Identify concrete failures | ||
| Length-limited fields fitting / checked | Record field, count, limit and source scope | ||
| Unselected channels or unnecessary content |
“Human action” counts depend on the interface: define a practical convention before the pair and retain it. Report both input effort and total instruction size if useful, but do not mistake an automatically exported long brief for extra employer typing.
First-output field and safety check
| Check | v1.4: pass / fail / N/A | Formula: pass / fail / N/A | Evidence |
|---|---|---|---|
| Exact selected channel/format delivered | |||
| SEEK title, summary, public pay when selected, supported selling points and body separated | |||
| LinkedIn Jobs fields distinct from a feed-post composer | |||
| Dropdown/settings notes outside public copy | |||
| SEEK uses appropriate platform application route; social has supplied destination | |||
| Missing hours or needed application route keeps affected output in draft | |||
| No invented pay, benefits, contacts, flexibility or progression | |||
| Essential versus desirable/trainable requirements preserved | |||
| Material conditions, contradictions and outstanding checks visible | |||
| No unsupported legal-compliance or recruitment guarantee | |||
| Style stays warm/clear without unsupported promises |
For SEEK's documented API text fields, check title ≤80, public pay label ≤50, summary ≤150, each selling point ≤80 and body ≤15,000 characters including markup. These documented limits do not establish that a signed-in native form was observed. Check other constraints against the recorded publishing route; do not import LinkedIn API limits into a feed post or invent a universal Facebook limit.
Repeatable scenarios
- Short facts with gaps: omit hours and a required social destination. Check draft labels, visible gaps and whether the AI asks only necessary questions. A SEEK-only run does not need an external destination.
- Complete existing ad pasted: include all factual details in the ad with sparse form answers. Check reuse, no repeated full questionnaire, and no lost conditions. Repeat with one deliberate conflicting form value; require a visible conflict rather than silent resolution.
- SEEK plus social: provide a social application email/link. Check SEEK uses its configured Apply route while Facebook/feed/short copy retains the supplied social destination.
- Exact channel selection: select one channel, then a combination. Check no unwanted outputs and no missing selected ones.
- Optional SEEK selling points: test an unknown/unsupported product and a product confirmed to support three points. Check product caveats, separate fields and per-field limits.
- LinkedIn format: run Jobs listing and feed post separately. Check structured listing fields/setup notes versus a single post block and appropriate application route.
- Style and factual boundaries: select joyful responsible abundance with limited benefits and incomplete pay checks. Check truthful warmth and clear agreements without invented perks, guarantees or a declaration of compliance.
Use the same facts for both methods in each scenario. If a method requires extra answers, record them and supply identical factual answers when comparable; count the effort. These are proposed tests, not completed trials.
Ease and outcome rubric
Score each dimension 1–5; use 2 or 4 for intermediate results.
| Dimension | 1 | 3 | 5 |
|---|---|---|---|
| Ease of starting | Confusing or cannot start | Some guidance/rework needed | Clear first step; can proceed unaided |
| Input effort | Repeated or lengthy entry | Moderate entry/questions | Mostly short answers; facts reused |
| Field readiness | Generic text needs major splitting | Some splitting or trimming | Required paste blocks and setup notes clear |
| Factual trust | Material invention/omission | Minor corrections needed | Facts/uncertainty preserved accurately |
| Candidate usefulness | Key decision facts hard to find | Mostly clear | Conditions, opportunity and application path clear |
| Voice | Hype, pressure or bland filler | Usable with editing | Natural, specific and appropriate |
A high average must not conceal a critical error. Mark not acceptable to publish yet whenever a material falsehood, missing application path, unresolved key condition or unsafe compliance claim remains.
Final judgement: ___
Concrete evidence of improvement or regression: ___
Changes recommended before employer trials: ___
Limitations (number of runs, model changes, account/form access, timing variability): ___