The same eight numbers, two charts — one you can update, one you can only redraw.
The full script, word for word — 713 words, about 3 minutes to read. Current as at August 2026. AI tools change quickly; if something looks different when you try it, check the product's own help pages.
Hi, I'm Mel. A chart can be right and still be the wrong object. Gemini made this one from eight figures and displayed every value correctly. Good result. The output is an image, while the chart beside it is tied to a table. One can be inspected and updated. The other must be checked again or remade. This video is about choosing the visual structure, and the tool, that lets the source stay in control.
Diagram, infographic and chart are three different jobs. A diagram shows how something works: a process, structure or relationship. An infographic arranges facts so one message lands quickly. A chart encodes values, so its scale, labels and source govern whether it can be trusted. They can appear together, but they fail differently. Start by naming the job. That decision tells you what has to remain controllable.
We gave Gemini four weeks of adult and child bookings and asked for a grouped bar chart. The first result passed: eight supplied values, eight correct labels, sensible axes and a readable legend. That is a useful improvement, and it changes the warning. A generated chart can be correct. This output is still a picture. The PNG does not show a source cell, formula or update path behind any bar.
In Canva, the same figures sit in a table beside the chart. Change a value and the bar follows it. The finished chart can still use the Fernhollow colours, labels and type. Styling has moved to the design layer; the values remain controlled by data. Use this route when a chart will be checked, revised or repeated. Generate the picture only when a static illustration is enough, then verify every displayed value against the source.
Structure is where a general assistant earns its place. We supplied a paragraph about inspecting a lion enclosure. Gemini found one decision and both paths: open the tour when safe; otherwise pause, update bookings, notify guests, obtain a manager review, then reopen or cancel. It added no unsupported step. That gives us a strong first structure. An operating owner must still confirm the decision, the order and every exception before it becomes an instruction.
A visual can be accurate and still fail because it carries too much. Decide the sentence it must communicate, then decide what the eye should see first. In this infographic, the planning cycle is the first look and the supporting stages sit around it. If every box, colour and number asks for equal attention, the viewer has to solve the hierarchy before they can understand the point.
Choose the route by the control the job needs. A fast AI visual tool buys speed. A design tool makes the result fit the page or campaign around it. A data-visualisation tool keeps the numbers connected and can support refresh and interaction. A formal diagramming tool gives documented procedures standard shapes and accountable connections. A text-based diagram language keeps the visual beside the document or code that defines it. The choice is speed, fit, data, rigour or control. There is no useful ranking without the job.
Now ask how often this object must work. For one file and one question, use a one-off answer. Upload the source, ask the question and check the result. Build a reusable calculator or scorecard when a bounded group will explore the same logic repeatedly. A governed dashboard belongs further to the right: data refreshes, definitions are agreed, permissions control access and an owner keeps it working. Start on the left. Move right only when the operating need earns the maintenance, governance and cost.
Before sharing, check four things. Can the viewer see or reach the source? Do the values, labels and scale agree with it? Does every connection in the diagram reflect the documented process? Is any material caveat close to the visual rather than hidden in your notes? A polished visual invites confidence. Give that confidence evidence it can survive.
Use AI to propose structure and accelerate the first visual. Keep the source where you can inspect it. Draw numbers in something that understands data, and draw governed processes in something built for decisions and connections. One message, one first look, and the smallest operating shape that solves the job. Then verify it before another person acts on it.