R0021/2026-03-25/Q003
Query: What specific, measurable guidance do the major AI vendors (OpenAI, Anthropic, Google, Microsoft) provide in their official prompt engineering documentation? What percentage of their recommendations are quantifiable versus subjective?
BLUF: Approximately 85-90% of vendor prompt engineering recommendations are subjective/qualitative. Across ~25 distinct recommendations from all four vendors, only 3-4 include quantifiable criteria. Microsoft explicitly calls prompting "more of an art than a science." No vendor provides engineering-grade specifications with testable success criteria.
Answer: H2 (Predominantly subjective) · Confidence: High
Summary
| Entity |
Description |
| Query Definition |
Question as received, clarified, ambiguities, sub-questions |
| Assessment |
Full analytical product |
| ACH Matrix |
Evidence × hypotheses diagnosticity analysis |
| Self-Audit |
ROBIS-adapted 4-domain process audit |
Hypotheses
| ID |
Statement |
Status |
| H1 |
Vendor guidance is predominantly quantifiable |
Eliminated |
| H2 |
Vendor guidance is predominantly subjective |
Supported |
| H3 |
Mixed quantifiable and subjective |
Partially supported |
Vendor Comparison
| Vendor |
Recommendations |
Quantifiable |
Subjective |
Notable |
| OpenAI |
6 strategies |
0 |
6 |
"Empirical discipline" |
| Anthropic |
7 techniques |
1 (semi) |
6 |
XML tags structural but unmeasured |
| Google |
7 strategies |
3 |
4 |
Temperature=1.0; ~21-word average |
| Microsoft |
5 practices |
0 |
5 |
"More art than science" |
| Total |
~25 |
~4 (16%) |
~21 (84%) |
|
Searches
| ID |
Target |
Type |
Outcome |
| S01 |
OpenAI documentation |
WebSearch |
2 selected, 1 rejected |
| S02 |
Anthropic documentation |
WebSearch |
2 selected, 1 rejected |
| S03 |
Google documentation |
WebSearch |
2 selected, 1 rejected |
Sources
Revisit Triggers
- Any vendor publishes quantifiable prompt engineering specifications with testable success criteria
- Industry consortium or standards body publishes formal prompt engineering standards