Assistant 1.30.4 can compare real products and calculate quantities from verified catalogue facts. These functions are deliberately stricter than ordinary chat: when compatible measurements or relationships are missing, the safe result is an explanation—not an invented number.

Configure measured comparisons

Open Settings → Appearance → Measured product comparisons. Enable the feature, then choose the default metric:

• Automatic
• Area or volume
• Length, width or height
• Weight
• Numeric attribute

Set Number of options for requests asking for several large or small products. A true superlative such as “largest” returns one verified product.

Enable variation measurements only when variations contain trustworthy, comparable data. Add Comparable attribute names and Comparable custom fields only for numeric fields that use stable units. Assistant normalises common metric and imperial units, but it does not turn a marketing label such as “XL” into an unverified dimension.

What a verified comparison uses

Side-by-side results are rebuilt from live WooCommerce facts and can include price, availability, SKU, dimensions and product attributes. The selected products must be resolvable in the current conversation; a comparison should not reuse unrelated products from an earlier task.

Test:

• two explicitly named products;
• “largest/smallest/heaviest/cheapest” inside one category;
• products with different compatible units;
• a product missing the requested measurement;
• a variation whose measurement differs from its parent.

When a value is missing, correct WooCommerce or the configured custom field. Do not add the missing number to Assistant instructions.

Deterministic calculation engine

The calculation engine distinguishes three evidence types:

• SOURCE_VERIFIED: a product or trusted source supplies the fact;
• USER_PROVIDED: the visitor supplies a dimension or target quantity;
• CALCULATED: Assistant applies a traceable formula to those values.

It supports compatible conversions and relationships for area, length, volume, liquid volume, mass, item counts and packaging. Examples include square metres per unit, units per package, square metres per litre, litres per square metre, kilograms per square metre, metres per roll and package sizes.

Discrete products are rounded up when the calculation asks how many units or packages are required. Conflicting verified coverage facts stop the calculation. A sheet-coverage fact cannot answer a screw-count request unless a separate verified relationship connects sheets and screws.

Safe two-hop calculations are possible only when both relationships exist—for example target area → panels required → fasteners per panel.

Prepare catalogue facts for calculations

1. Use clear attribute or custom-field labels that state both quantity and unit.
2. Keep shipping dimensions separate from usable coverage. Roll coverage is derived only when material width and length are explicitly labelled.
3. Avoid storing multiple contradictory coverage values in description, attributes and custom fields.
4. Include variation facts when the chosen variation changes package size or coverage.
5. Reindex after changing product knowledge fields.

Configure visual search

Visual search requires a saved API key and a selected image-capable model. Gemini and OpenAI support vision; Claude supports vision; Grok requires one of the configured vision models. If no usable provider path exists, the front-end image-upload control is hidden.

The uploaded image is sent server-side to the configured vision provider to identify a product type, then Assistant matches that description against the real catalogue. The provider does not create a new product and should not be trusted as the final source of price, stock or compatibility.

Test with:

• a clear photo matching a catalogue category;
• a visually similar but unavailable product;
• a photo containing several objects;
• an unsupported file or oversized image;
• a product whose appearance is similar but technical specifications differ.

Success means the returned cards are real, relevant catalogue products and the explanation does not claim unseen specifications.

Troubleshooting

• Compare request produces prose only: name two resolvable products or establish them in the immediately preceding turn.
• Largest/smallest is wrong: check the selected metric, unit, variation data and numeric attribute mapping.
• Calculation refuses to answer: provide the missing target dimension or add a verified product relationship; refusal is correct when evidence is incomplete.
• Wrong output unit: make the requested output explicit and check catalogue labels.
• Image upload is absent: configure and test an image-capable provider/model.
• Visual result is semantically right but cards are wrong: rebuild the product index and test the generated product-type phrase against Search test.
• A result uses stale facts after navigation: start a new conversation and verify current WooCommerce data and cache layers.

Verification checklist

• Every comparison value exists in live product or variation data.
• Common units normalise correctly and incompatible units do not compare.
• Discrete quantities round up and formulas are explainable.
• Missing or conflicting facts cause a safe stop.
• Visual search returns only real catalogue cards.
• Image-derived descriptions never override live price, stock or technical data.