Free · no sign-up
Read the answer the way you would read the chart itself
A fluent reading and a correct one look identical on a screen. These are the checks that separate them, and the questions worth asking next.
The reading is in front of you. It is organised, confident, and about you, and none of that is evidence that it is right. The prompt did the difficult half: positions to the arc-minute, the tightest aspects with their orbs, and a page of instructions on what it may not do.
This page is the other half: what to do with the answer. Which parts of it carry weight, which questions turn a broad reading into a specific one, and where even a well-instructed model slips. The prompt itself comes from the natal chart prompt page.
Why a fluent answer still has to be checked
A model writes with the same confidence whether quoting your chart or filling a gap from memory, and the passages most likely to be wrong often read best: an invented contact makes a tidier paragraph than a real one at an awkward orb.
The prompt anticipates that: positions may not be recalculated, the aspect list is complete, every placement interpreted must be quoted with its degree, and a line that would fit anyone counts as a failure. Those rules cut the error rate rather than removing it, and each leaves something to check.
Get the prompt first
This page is about reading the answer. The prompt itself — your chart computed to the arc-minute, with the instructions the model needs — comes from the natal chart prompt page.
Open the natal chart promptHow to get your reading
Paste the block whole, and nothing under it
Send the copied text as the first message of a new chat, on its own: it ends with instructions about structure, tone and language, which a question added underneath competes with.
Read the shape before the content
The answer should arrive under thematic headings, close with the three most important points, then offer three questions to ask next. That shape is requested in the prompt, so a wall of text means part of the block never arrived.
Ask the follow-up questions it hands you
The three suggestions exist so the conversation continues while the chart is still in context. Take one, then ask your own: which follow-up questions repay the effort is the next section.
Verify three claims, one from each layer
Pick a placement, an aspect and a claim about the balance, and find the line each came from. The balance counts the ten Sun to Pluto bodies only. Three lookups take two minutes and say more than a second reading.
Keep what survived
Keep the claims that cite a placement, name an orb, and are concrete enough to disagree with. The rest were never about your chart.
What a well-formed answer does
It quotes. Each placement arrives with its degree and sign as printed, in the form Sun at 20 degrees 39 minutes of Leo in house 12, and each aspect carries the orb from its row. It works through the numbered aspects; a wider one from the reference block earns at most a clause.
When an element or modality reaches forty percent or more, the answer names that group and quotes the figure as printed. The angles carry no aspect rows, so they are read as positions; the nodes can appear as rows, and are read as an axis by sign and house besides. And it should be willing to be unwelcome: the prompt asks for the costs too.
The second and third question
A first reading is broad by design: a whole chart in two or three pages, so no theme gets depth.
Name one placement with its degree and house, and ask what it does in this chart rather than in general. Cite an aspect by its row number and orb, and ask what it costs rather than what it gives: asking by number keeps the model on data. Or name a house, its cusp sign and the bodies inside it.
Nothing else repays asking: transits, progressions, solar returns, fixed stars and a partner's chart were never in the block.
Checking the answer against the wheel
Checking is quick because the model must cite the placement it reads. Where a paragraph asserts something with no degree behind it, ask which line it came from: a claim that cannot name its source dissolves.
An invented aspect is the failure to watch hardest for. The prompt prints one numbered list of the chart's tightest contacts, at most eight, each with its orb; wider ones may sit under it marked reference only. A pair in neither is not in contact, however close the degrees look, and two bodies sharing a sign are a grouping, not a conjunction. No row, no aspect, and a hedge such as visible from the overall picture is the tell.
A recomputed position is subtler. Compare the degree quoted against the line in the block, and both against the natal chart calculator: a Mercury that has moved a degree is working from memory.
Where models still slip
Generic lines come back under pressure. Sensitive underneath but composed on the surface fits almost anyone, which is why it feels accurate; it turns up where the model has run out of chart.
House systems get mixed. The block names the system and prints twelve cusps, but a model steeped in whole-sign astrology puts a body in the house its sign suggests, not the one on its own line.
A wide orb gets read as a tight one: rows run from the most exact downwards, and a contact several degrees off does not carry a section of your life.
An unknown birth time gets filled in silently. Without a time the prompt drops the houses and the angles rather than estimating them, so a rising sign in the answer came from the model, not the chart.
What checking cannot settle
Verification catches an answer that contradicts the data. It cannot make an interpretation true: two astrologers reading the same square write different paragraphs, both inside the tradition.
An exact chart is not a forecast either. What comes back describes tendencies, not events, and settles no medical, legal or financial question. The last check is yours: a well-sourced claim you do not recognise is still one you may set down.
Questions
Compare what each was given, not which reads better. If both started from the same computed block, the difference is interpretation and neither is authoritative. If one came from a date typed into a chat, its positions were guesses.
Look for its row in the text you pasted. The numbered list is the whole set the reading may interpret, and wider contacts sit below it marked reference only. A pair in neither does not aspect here.
Then it made one up. With the time unknown the prompt removes the houses and the angles rather than estimating them, and says not to infer them. Ask for the reading again without them.
Yes, once your corrections outnumber the content. A long thread keeps every earlier mistake in view, and a model defends what it has said. Paste the block into a new chat, narrow question first.
Not by itself. Recognition is what a generic statement is built to produce; a reading that never made you uncomfortable stayed on safe ground. Move a sentence into a stranger's reading: if it fits, it was never about you.
It helps a little. Asked to list every claim beside the placement it came from, it drops the unsourced ones. But a model that invented a contact will justify it twice, so do the lookups yourself.