An AI Report From Social Media Data

Rows you have already fetched in Ask AI can become a written report or one PNG. A report with no images is 50 credits. Each image you add costs 50 more, and you can add 0, 1, 2, or 3, so the report totals are 50, 100, 150, or 200 credits. An infographic is 50 credits at every shape. The Generate button shows that total before anything is charged. Offering the card spends nothing. The charge starts when you press Generate. The model does not pick the price.

Those figures are the constants the button uses. A production walk-through of a finished PDF was not part of this article.

The rows have to be in the chat already. Fetching them is a different card, the plan you approve before the data is pulled. That flow is covered in how Ask AI fetches and how you approve the plan. This page starts after the rows exist. API credit rules are in the credits docs. Credit packs are on pricing.

What does the intake card ask before you spend?

The intake card is the set of questions for this file. It is not the plan card that fetched the rows.

A report card asks four things, and one of them can be hidden.

Who will read it. The options are Me or my team, A client, Leadership, and The public. The card opens on Me or my team unless the chat already chose.

What should it help decide. You get options written for this chat, or you type your own. That typed decision stops at 500 characters.

Real posts and quotes. This row appears when the stored data has comments, or an author name or handle, or when the chat already answered it. The options are With names and links, Without names, and Numbers only. If the row is absent, the report uses numbers only.

Images in the report. None, 1 image, 2 images, or 3 images. The hint says each image adds 50 credits, and the total on the button changes as you pick. With none selected, a report opens at 50 credits.

Under the rows, an optional note is labelled Anything else you want to share. It holds up to 1,000 characters. Use it for a deadline, a name to leave out, or a comparison you care about. It does not change the price by itself.

The button reads Generate for N credits, with N taken from the choices above. Cancel on that card spends nothing. Replying yes approves the card. Replying no cancels it. Any other message replaces the card.

The card also shows your balance. If the balance is short of the total, the line names both numbers: how many credits you have, and what this costs. Buying more is a separate step on pricing.

A card left open for more than 24 hours closes on its own. It says the card is more than 24 hours old, so it closed without charging anything. The action is Ask again, which asks for a new card and does not charge the old one.

If the stored data is thin, the same card says so before you pay. The rules for that note are below. The button on the note is Fetch more first.

Which of the four report jobs should you pick?

The report is one of four jobs. Each file starts with an answer, then tables, charts, quotes, recommendations, limits, and method.

On screen, Method is left off. Download PDF, Download Word, and Download Markdown keep the document's sections, so Method is in the file. Copy Markdown is there too.

Brand sentiment is the job for how people responded in the comments you already stored. If those rows hold no comments, the card says it cannot say how people feel, and it offers to fetch comments before you pay.

A creator shortlist is the job for accounts in the stored rows. If what you have is profiles, and those rows have no per-video views, likes, or comments, the card says it cannot show how their videos perform. Fetching recent videos is the fix it suggests.

A competitor comparison is the job for two or more sides already in the chat. It is ranked only when every side has at least 30 rows. Below that, the report does not rank them, and the card says ranking the sides would not be fair. One side is not a comparison: the card says it cannot compare anything.

General is the same document when the job is not one of those three. A short comment set still gets a thin-data note. Zero comments on a general job does not use the sentiment sentence. That sentence is reserved for the brand-sentiment job.

You pick the job by what you ask for in the chat. The card then shows the price for the images you chose, not a different price per job.

What does Checking figures change in the draft?

Once the first words exist, the report opens beside the chat on a wide screen, once per report. On a narrow screen that panel would cover the chat, including anything you are still typing, so it does not open on its own.

While the report is still being written, the panel can show an outline. That text is an unchecked draft. Leave it there until the check finishes.

The panel then switches to the label Checking figures. The progress line on the card uses a different string, Checking every figure. The panel note says every figure is being checked against your data, so a sentence may still change. When the check finishes, the checked text replaces the draft. A sentence you saw in the outline can be gone, or rewritten, in that step.

A check runs. It does not promise that every sentence is right. The header tells you to read the report before you forward it. Every figure comes from your data, and a model wrote the prose.

Figures are computed from the stored rows. Quotes and links are copied from stored records. A model does not type the figures, and it does not type the URLs. If a stored row is wrong, the report can repeat that wrong row. The check compares claims with those stored rows. It does not go back to the platform and pull the post again.

The card names the stage it is in. For a report those labels are Reading the data, Writing the report, Checking every figure, Drawing the images, and Saving the result. It does not draw a fake percent. If the wait runs long, the card says a long report can take up to 5 minutes. That sentence is the label on the card. It is not a time measured for this article.

When does the card warn that the data is thin?

The warning is on the intake card, before Generate charges you. You can still pay. You can also press Fetch more first and send the suggested follow-up, which asks for more rows before the report is written.

A note appears when the stored comments are fewer than 30, or when those comments come from fewer than 5 posts and the post count is known. The wording names those counts, then says that one post, or a few comments, can move the shares.

A brand-sentiment job with no comments does not guess a mood. The card says it holds no comments, so it cannot say how people feel. The suggested next message is to fetch comments on these posts before writing the report.

A comparison is not ranked unless every side has at least 30 rows. If the smaller side is under that, the card names the side and the count, and says ranking the sides would not be fair. If the chat only has one side, the card says it cannot compare anything, and it tells you to fetch the other side first. The report, if you generate it anyway, states that the sides were not ranked. It does not invent a winner to fill the page.

A creator shortlist built from profiles, with no per-video figures, says it cannot show how their videos perform. The count in that sentence is the number of profiles. The suggested fix is to fetch recent videos for those creators before writing the report.

One more case sits behind those. If the fetch that produced the rows stopped before it finished, and none of the notes above already fired, the card says the figures cover only part of what was asked. It suggests running the same plan again and finishing it before the report.

What is refunded if the report or the image fails?

The refund runs once. What comes back is the credits charged, minus the value delivered, minus anything already refunded. A second pass does not pay again.

For a report, delivered value is 50 if the written report was delivered, and 0 if it was not, plus 50 for each image that was placed. That sum is never higher than the charge. A report that fails outright delivers nothing, so the charge comes back.

Under that rule, three images are a 200 credit charge. If the written report is delivered and one image is placed, delivered value is 100, and 100 credits come back. If the written report is not delivered, its 50 is not kept, even when an image was placed. That example is the rule. It is not a receipt from a finished PDF.

A failed infographic image is not shown. Delivered value is 0, so all 50 credits come back. The same facts are drawn as a stat card, with the headline and the figures and no image. A failed image does not still cost 50 credits. An image that passes keeps the 50, because the charge and the delivered value are the same.

You can have three generations active at once. A fourth start is refused. The card says you already have 3 generations running, tells you to press Generate again once one has finished, and says nothing was charged.

Pressing Generate again on a row that has already started does not charge a second time. And an intake older than 24 hours never reaches the charge. It asks you to offer the card again.

Which PNG shape do you get, and can you edit it?

An infographic is one image of a finding from the same stored rows. The price stays 50 credits for every shape. The card opens on Tall post unless you already chose another.

You also choose what it should show, from options written for this chat or from your own words, and you can add the same optional note, up to 1,000 characters. Neither choice changes the 50 credits.

The download is a PNG. There is no editor on the image. There is no editor on a finished report either. The header has no editing toolbar, because editing is not built. If a sentence is wrong, the fix is to read it, then fetch different rows or ask for a new file. You do not patch the PDF in place.

While the image is drawn, the progress label for the check is Checking the figures on the image. That is the infographic's label. It is not the report panel's Checking figures. An image that fails that check is dropped. You get the stat card and the refund above, not a PNG with a number the check rejected.

What should you read before you forward the file?

Treat the file as an analyst's draft for a person to review. It does not replace an analyst. The header tells you to read it before you forward it, because a model wrote the prose around figures that came from your rows.

A report on thin data says so on the card before you pay, and the limits section of the file keeps the caveats that bear on the decision. A creator shortlist without per-video figures says it cannot show how the videos perform. A comparison without 30 rows on every side is not ranked. Those jobs are weak on thin data because they say they cannot show the thing you asked.

Three generations can run at once for your user. A fourth start is refused and does not charge. You press Generate again after one of the three has finished.

When you want the file, open Ask AI on a chat that already holds the rows, read the total on Generate, and read the PDF before you send it to a client.

What should you confirm before you press Generate?

How many credits does a report cost?

A report is 50 credits with no images, plus 50 credits for each image you add. You can add 0, 1, 2, or 3 images, so the totals are 50, 100, 150, or 200. An infographic is 50 credits at every shape. The Generate button shows the total for the choices on the card. Offering the card spends nothing. The charge starts when you press Generate. The model does not set the price.

Does a failed infographic still cost 50 credits?

No. A failed infographic image is not shown. All 50 credits come back, once, and the same facts are drawn as a stat card. An image that is delivered keeps the 50. A failed report is refunded the same way: credits charged minus value delivered. Delivered value is 50 if the written report was delivered, otherwise 0, plus 50 for each image that was placed, and never more than the charge.

Is the report on screen already checked?

Not while it is still writing. The side panel can show an outline during that time, and that draft is unchecked. The panel label Checking figures means the check is in progress. A sentence may still change. The checked text replaces the draft when that step finishes. A check runs. It does not promise that every sentence is right, so read the file before you forward it.

Can you edit the PDF or the PNG?

No. A finished report has no editing toolbar in the header, and the PNG has no editor either. Downloads are PDF, Word, and Markdown for a report, and PNG for an infographic. To change the file, fetch different rows or ask for a new one, and approve the new total on Generate before it runs.