Data Chat Guide

Data Chat Guide

Info
Ask a question about your marketing data in plain English and get an answer in about 25 seconds, built only from verified queries against your hub's data. This guide shows you where to find Data Chat, how to ask questions that get great answers, and how credits work, with worked examples you can try today.
Available on: Business, Grow, Pro, and Custom plans (alpha feature, enabled per hub)

What Is Data Chat?

Data Chat is a conversational way to explore the data in your hub. Type a question such as "How much did I spend on Google Ads last month, and how does that compare to the month before?" and Data Chat finds the right sources, runs the queries, checks every figure against the results, and writes the answer, often with a chart.

Three things make Data Chat different from a general AI chatbot:
  • Every number comes from your data. Each figure in an answer comes from a real query against your connected sources, never from the AI model's memory. If a figure can't be verified, Data Chat tells you so instead of guessing, and you aren't charged.
  • It's read-only. Data Chat reads your data and never changes it. It can't edit campaigns, budgets, datasets, or settings.
  • It only sees your hub. Data Chat works with the sources connected to the hub you're signed into, and nothing else.

Who Can Use Data Chat?

Notes
Data Chat is currently an alpha feature. It's switched on for hubs that have alpha features enabled, and the Data Chat tab shows an ALPHA badge. If you don't see the tab, contact our support team and ask us to enable it for your hub.
For Data Chat to answer a question, three things need to be true:
  • Your hub's subscription is active (live, in trial, or in its final paid period).
  • Your plan includes AI Insights credits. Data Chat uses the same monthly credits as AI Insights. The Free plan doesn't include AI Insights credits, so Data Chat can open on a Free plan but can't answer.
  • You're a member of the hub. Data Chat only reads the sources connected to the hub you're signed into.
Alpha means Data Chat works and is checked every day, and that its wording, limits, and pricing can still change. We'll let you know before anything about credits changes.

Where to Find Data Chat

There are three ways into Data Chat, and all three open the same conversations.

The Data Chat Tab

  1. Select Data Insights in the left menu of your hub.
  2. Choose the Data Chat tab on the AI Insights page.
  3. Type your question in the Ask about your data box at the bottom and press Enter (or click the arrow).
Select Chat History to see your hub's conversations, newest first. From there you can start a fresh conversation with + New chat or find an earlier one with Search your chats.

Ask Your Data, from an AI Insight

Open any AI insight and click Ask your data. A Data Chat panel opens beside the insight, already knowing the insight's title, date range, and chart, so you can ask "why did this happen?" or "break this down by campaign" without repeating the context. A chat opened from a dataset insight stays focused on that dataset. Click Open in Data Insights to continue the same conversation in the full Data Chat tab.
Every conversation has its own web address. Copy the address bar while a chat is open and send it to a teammate in your hub, and they'll open the same conversation.

How to Ask a Question

Ask the way you'd ask an analyst on your team: name the source, the metric, and the time window. Data Chat understands relative dates such as "last month," "the last 30 days," or "July," and every answer tells you the exact dates it used.

You want
Ask something like
A total
"How much did I spend on Google Ads last month?"
A comparison
"Compare this month to last month for each of my ad platforms: spend and conversions."
A ranking
"Which Facebook campaigns had the highest cost per lead in August?"
A breakdown
"Break down Shopify revenue by product for the last 30 days."
A trend
"How has my Google Ads cost per click moved month by month this year?"
A pattern
"Which day of the week gives me the cheapest clicks?"
Unusual days
"Were there any unusual days in my TikTok Ads spend this quarter?"
An outlook
"Am I on pace to beat last month's Shopify revenue?"
A cause
"What drove the change in my Facebook cost per lead between July and August?"
Across sources
"How much did I spend on ads versus how much revenue Shopify made in the last 30 days?"
One client or account
"For [client name], how did Facebook Ads cost per click move quarter by quarter over the last 12 months?"
The health of your sources
"Which of my accounts have expired tokens or stopped syncing?"
Your datasets
"How many datasets do I have, and which sources feed them?"

After you press send, you'll see a short status line for each stage ("Analyzing your question...", "Querying your data...", "Calculating statistics...", and "Drawing your chart..." when there's a chart), plus small chips naming the queries as they run. A typical answer takes about 25 seconds; a question that spans many accounts or sources can take up to a minute and a half.

Worked Examples

Example 1: A Total, Then a Follow-Up

  1. Ask: "How much did I spend on Google Ads last month?"
  2. Data Chat answers with the exact dates it used (for example, "For September 1 to 30, 2026"), the total spend, and how many of your Google Ads accounts contributed to the total. Under the answer, chips such as pma_query_custom · 2 queries show which queries ran, and a credits line such as "Used 2 credits · 48 remaining this month" shows what the answer cost.
  3. Follow up: "Now break that down by campaign." Data Chat runs new queries for the same month and usually adds a ranking chart. Because it needed new data, this follow-up costs 2 credits.
  4. Follow up: "Which dates did you use?" This is a question about the conversation itself, so Data Chat answers from what it already found, and it's free.

Example 2: Spend Without Conversions in an Agency Hub

  1. Ask: "Which ad accounts spent money in the last 7 days but have zero conversions tracked?"
  2. Data Chat checks each connected ad platform (for example, Facebook Ads and Google Ads), states the exact 7-day window, and lists the accounts that spent but recorded no conversions, largest spenders first, often with a chart of spend by account.
  3. Spend stays in each account's own currency. When accounts bill in different currencies, Data Chat lists them separately and reminds you not to add them together.
  4. Read the coverage caveat at the end. It tells you how many accounts returned data for the window and which were left out, for example accounts with expired or revoked tokens. Reconnect those sources and ask again to include them.

Example 3: Digging Into an AI Insight

  1. Open an AI insight for one of your datasets and click Ask your data.
  2. Ask: "What drove that change?" Data Chat already knows the insight's dataset and date range, so you don't need to restate them.
  3. Ask: "Break this down by campaign." The answer stays scoped to that dataset.
  4. Click Open in Data Insights to keep working in the full Data Chat tab.

Example 4: When Data Chat Asks You to Narrow It Down

  1. Ask: "How are we doing?"
  2. Data Chat replies with a clarifying question, asking which source, account, or date range you mean. Clarifying questions are free.
  3. Ask: "List every order in the last two years." Each query returns at most 100 rows, so Data Chat works best with totals, rankings, or per-period figures. Try "Show my monthly Shopify orders and revenue for the last two years" instead.

Reading a Data Chat Answer

Every answer states the dates it covers, the figures, and the caveats that apply to them. The caveats tell you how much of your data the figures rest on, so it's worth reading them before you share a number.
  • Dates: "For August 12 to September 10, 2026" is the exact window the queries ran on, never an approximation.
  • Coverage: Lines such as "7 of 8 Shopify stores returned data" or "21 accounts could not be reached" tell you which accounts are in the total and which aren't.
  • Currency: Each account's figures stay in its own billing currency. When accounts use different currencies, the answer keeps them apart instead of summing them.
  • Data through: "Data runs through September 8" means that's the last day your source has synced. The current day is usually partial.
  • The chart: One chart appears when it helps (a ranking, a trend, or a comparison), and its points come straight from the query results. You can ask for the shape you want, such as "show me a monthly trend" or "as a ranking."
  • Source chips: Chips under the answer, such as pma_list_data_sources · 1 queries or pma_query_custom · 3 queries, name the tools that ran and how many queries each made. Think of them as the answer's receipt.
  • The credits line: Charged answers show the credits used and your remaining balance for the month. Free answers show no charge.
Notes
The figures in an answer are verified against your data, but the wording and reasoning around them are written by AI. Treat an answer like a first draft from a sharp analyst: reliable numbers, worth a second read before a big decision. That's why each conversation reminds you that answers are AI-generated and to double-check anything important.

Follow-Up Questions and Conversations

A conversation remembers its last six exchanges, so you can build on an answer instead of restating it: "break that down by campaign," "and the month before?" and "which of those had zero conversions?" all work.
  • Follow-ups that need new data run new queries and cost 2 credits, like any answer.
  • Questions about the conversation itself ("What did you mean by that?", "Which dates did you use?", "Thanks, that helps") are answered from earlier replies, for free.
  • Start a new chat when you change topic. Because the conversation remembers recent exchanges, an old thread can pull a new question toward an old context.
  • Conversations are shared across your hub. Every member of your hub can see, search, open, and continue every chat in the list. There's no private chat today.
  • Deleting a conversation: You can't delete a conversation from the hub today. Contact our support team if one needs to go.

How Data Chat Credits Work

Each answered question costs 2 AI Insights credits. Everything else is free.

Outcome
Credits
An answer with verified figures
2
A follow-up about the conversation itself
0
A clarifying question back to you
0
A reply that your question is outside what Data Chat can answer
0
"Your connected data doesn't contain the information needed to answer this question."
0
"I couldn't verify the numbers behind this answer..."
0
Any error
0
Sending the same question twice by accident
0 for the duplicate

Data Chat draws from the same monthly pool of AI Insights credits as AI Insights in Data Builder. Here's what each plan includes:

Plan
AI Insights credits per month
Data Chat answers if used only on chat
Free
0
0
Business
50
25
Grow
100
50
Pro
100
50
Custom
Set by your agreement
Half your monthly allowance
  • Your credits reset on the first day of each calendar month. Unused credits don't roll over.
  • When the month's credits are used up, Data Chat replies "Monthly insights limit reached. Please upgrade your plan to continue using AI Insights." and nothing is charged.
  • Your remaining credits show in the AI Insights box at the top of your hub and on every charged answer.
  • You're only charged once. Credits are charged after the queries succeed and before the answer is written; if anything fails after that point, the charge is refunded right away. Sending a question twice, refreshing mid-answer, or losing your connection doesn't charge you twice.

What Data Chat Can and Can't Do

Data Chat Can

  • Total, average, and count any metric your sources report, for any date range inside your stored history.
  • Compare two periods, two sources, or two accounts, and say what changed.
  • Rank campaigns, ad groups, keywords, products, pages, stores, or accounts by any metric.
  • Break a metric down by a dimension such as campaign, device, country, product, channel, or day of week.
  • Calculate rates such as cost per click, cost per lead, return on ad spend, and average order value, and compare them across periods.
  • Show a trend day by day, week by week, or month by month.
  • Spot weekly patterns, the day a metric's level shifted, and days that are unusual once the weekly pattern is removed.
  • Project the rest of a month from the trend so far, with a range, and tell you whether you're on pace.
  • Explain what drove a change in a rate: better or worse performance, or volume shifting between segments.
  • Report which of your accounts returned data, which couldn't be reached, and which have expired credentials.
  • List your datasets and sources, and read a dataset's rows.
  • Answer in the language you ask in. Source and metric names stay as your sources report them.

Data Chat Can't

  • Change anything. It can't create or edit datasets, pause campaigns, move budgets, or touch settings.
  • Use data you haven't connected. No industry benchmarks, competitor figures, web lookups, or general marketing advice.
  • See dates before your stored history begins or after your last data sync.
  • Return more than 100 rows from one query, or read a long date range row by row.
  • See another hub's data, even if you belong to both hubs.
  • Remember anything across conversations, or run on a schedule and email you.
  • Work outside the hub. Data Chat isn't available in Google Sheets, Microsoft Excel, or Data Studio yet; it lives on the Data Insights page and inside AI insights.

Your Data in Data Chat Answers

Data Chat queries the data PMA has already synced from your sources, the same data behind your reports. It doesn't call Google, Meta, or Shopify live, so an answer is as fresh as your latest data sync, and its figures match what you see in your reports.
  • Stored history: Data Chat can only see the history PMA has filled in (backfilled) for an account. For example, if an account's data starts in October 2025, Data Chat can't answer about September 2025. If you ask about an earlier month, it tells you the window is outside the stored range (for free) instead of answering from nothing. To look further back, backfill a longer date range for the account (Action > Backfill data range; see Manage and Configure Data Sources).
  • Several accounts: By default, a source's figure adds up every synced account under it. Name an account or a client to focus the question, or ask for a ranking by account. The answer always says how many accounts contributed.
  • Accounts that can't be reached: An account whose sign-in permission (token) has expired or been revoked, or that has a gap in its data sync, is reported as "could not be reached" and left out of the total. Reconnecting the source fixes it.
  • Large hubs: In a hub with a hundred or more accounts, name the client or account you mean. Data Chat matches names to accounts and ranks per account.
  • Time zones: Dates follow each source's own reporting day, the same convention your reports use.

When Data Chat Says No

Every refusal is free and tells you what to do next. A refusal is a result, not a failure: when Data Chat can't stand behind a number, it says so rather than producing one that only looks right.

What you see
Why
What to do
"This chat answers questions about your connected marketing data. Try asking about your campaigns, spend, or conversions."
The message wasn't about your data (general knowledge, coding help, advice, small talk).
Ask about a source, a metric, and a time window.
"Could you narrow that question down? Naming the source, account, or date range you mean will get you a precise answer."
The question was about your data but too open to query.
Add the source, the account, or the dates.
A clarifying question in Data Chat's own words
Your question could be read two ways that lead to different queries.
Answer it; the next turn uses your answer.
"Your connected data doesn't contain the information needed to answer this question."
The queries came back empty, or the metric doesn't exist in your sources. The answer usually names what's missing.
Check that the source is connected and synced, or ask about a metric the source reports.
An explanation that the window is outside the stored range
The dates are before your history begins or after your last data sync.
Ask inside the range the answer names, or backfill the account.
"I couldn't verify the numbers behind this answer, so I'm not showing them. Try a narrower question, like a shorter date range or a single platform."
A figure in the draft answer couldn't be matched to the query results, twice.
Ask a narrower question: one source, a shorter window, one metric.
"Data Chat is unavailable because this hub's subscription is inactive. Update your billing and try again."
The hub's subscription has lapsed.
Update billing; Data Chat works again right away.
"Monthly insights limit reached. Please upgrade your plan to continue using AI Insights."
The hub has used its AI Insights credits for the month.
Wait for the first of the month, or upgrade your plan.
"There was an error generating this answer." and a Try again button
Something failed on our side, including a question that ran past the five-minute limit.
Click Try again. If it repeats, narrow the question and let support know.

Limits and Important Considerations

Limit
Value
What it means for you
Rows per query
100
Ask for totals, rankings, and per-period figures, not raw row lists.
Size per query result
64 KB
Very wide breakdowns are trimmed, and the answer says when a result was cut.
History used for patterns, level shifts, and forecasts
365 days
Longer histories are compared period by period instead.
Time per question
5 minutes
Past this, the turn ends with an error and no charge.
Conversation memory
Last 6 exchanges
Older turns aren't read; start a new chat for a new topic.
Conversations in the list
50 most recent
Older chats stay reachable by their link.
Questions at a time
One per conversation
Wait for the answer before sending the next question.
  • If the status line and query chips stop moving for more than two minutes, something is wrong. The turn will end with an error you can retry, and retrying doesn't charge twice.
  • Generative AI can make mistakes in its wording and reasoning. The figures are verified against your data, but double-check anything important before acting on it.

Privacy and Security

Data Chat runs inside your hub. Every query is limited to your hub, every read is checked against your hub membership, and nothing Data Chat produces is written back into your data.
  • Who can see a conversation: Every member of your hub. Conversations are stored with your hub, like insights, and our support team may review them to help you and to improve answer quality.
  • Where your data goes: Queries run against PMA's own data warehouse, the same place your reports come from. No new connection to your sources is opened for a chat.
  • What the AI models receive: Your question, the conversation's recent turns, and the query results needed to answer (aggregated figures and the names of the accounts, campaigns, and products in them). Never credentials, tokens, or raw exports. Your data and questions aren't used to train AI models.
  • What's stored: Your question, the answer, the chart, a record of which queries ran, the verification result, and the credits charged. Conversations are kept for the life of your hub.
  • Sample data: Hubs on sample data get sample answers; no real customer data is mixed in.

Tips for Better Answers

  1. Name the source. "Google Ads spend" beats "ad spend" when you have five ad platforms connected.
  2. Name the window. "August" or "the last 30 days" gets you exact dates; "recently" gets you a clarifying question.
  3. Ask for one thing per question. Three questions in one message make a longer, slower answer and a bigger chance of a verification miss.
  4. Prefer totals and rankings over lists. "Top 10 campaigns by spend" answers in one query; "list all campaigns" hits the 100-row limit.
  5. For long histories, ask by period. "Quarter by quarter over the last two years" works; "every day for two years" doesn't.
  6. Name the client or account in an agency hub. Otherwise the total spans every account you manage.
  7. Ask "why" as a follow-up. Get the figure first, then ask "what drove that?"
  8. Use the free follow-ups. "Which dates did you use?" and "What did you mean by that?" cost nothing.
  9. Read the coverage caveat before quoting a total. "1 of 51 accounts returned data" changes what the number means.
  10. Start a new chat when you change topic.

Troubleshooting Data Chat

Symptom
Likely cause
Fix
No Data Chat tab on the Data Insights page
Alpha features aren't enabled for your hub.
Contact our support team to enable them.
Every question says the subscription is inactive
The hub's billing has lapsed.
Update billing in your hub's Billing settings; no wait needed.
"Monthly insights limit reached" on your first question
You're on the Free plan, or this month's credits were already used on AI Insights.
Check the AI Insights credit box at the top of your hub; upgrade, or wait for the 1st.
A total covers fewer accounts than you have
Expired tokens or revoked access on the missing accounts.
Ask "Which of my accounts have expired tokens?", then reconnect those sources.
The answer says the window is outside the stored range
That account's history wasn't backfilled that far, or the account stopped syncing.
Ask inside the named range, or backfill the account.
"I couldn't verify the numbers" on a broad question
The draft cited a figure the queries didn't return, twice.
Narrow to one source, one metric, or a shorter window and ask again.
A figure looks off
Usually coverage or currency.
Read the caveats, or ask "How many accounts contributed, and in which currency?" (free), then ask again more narrowly.
Nothing happens for more than two minutes
A query is stuck or the connection dropped.
Wait for the error and click Try again; retrying doesn't charge twice.
A follow-up ignored the previous answer
The conversation is past six exchanges, or you're in a different chat.
Restate the source and window, or start a new chat.
A teammate can't open your conversation link
They aren't a member of the hub, or the link was cut off.
Invite them to your hub, or resend the full address.

Frequently Asked Questions

Is Data Chat using live data from Google or Meta?

No. It uses the data PMA has synced into your hub, the same data behind your reports. That's why the figures match your reports and why each answer names the last synced day.

Can I trust the numbers?

Every number is checked against the query results before you see it, and a number that fails the check is held back. Read the coverage and currency caveats too: a correct number based on 1 of 51 accounts still describes only 1 account.

Why was I charged for an answer I didn't find useful?

A verified answer costs 2 credits whether or not it was the answer you hoped for. Follow-ups about the conversation, clarifying questions, and refusals are free. If an answer's figures are wrong, send us the conversation link and we'll review the charge.

Do my teammates see my chats?

Yes. Conversations belong to the hub, and every hub member can read and continue them.

Does Data Chat learn from my data or my questions?

No. Nothing is used to train AI models, and each conversation remembers only its own last six exchanges.

Can Data Chat create a dataset or change a campaign for me?

No. Data Chat is read-only by design. To build datasets, use Data Builder.

Why does the same question sometimes take 20 seconds and sometimes 90?

The time is spent in your data. A single-source question is one or two queries; a question across a large agency hub can run dozens of queries in parallel and waits for the slowest one.

Questions and Feedback?

Data Chat is an alpha feature, and every report we receive is read by the team that builds it. Tell us when an answer is wrong, when a refusal should have been an answer, or when something felt off. When you submit a ticket, please include:
  • The conversation link (copy the address bar while the chat is open). It lets us replay exactly what ran.
  • The question you asked and what you expected instead.
  • For a wrong figure, where the right figure comes from: a report, a screenshot from the source platform, or a dataset.
  • Whether credits were charged.
We'll reproduce the issue, fix the cause when it's ours, and let you know what changed.

    • Related Articles

    • CallRail Data Connector User Guide

      CallRail turns your phone calls, texts, and form submissions into marketing data you can report on. With the Power My Analytics CallRail connector, you can pull your call tracking, attribution, and Conversation Intelligence data into your favorite ...

    • Bing Webmaster Tools Data Connector User Guide

      Bing Webmaster Tools is Microsoft's free service for monitoring how your website performs in Bing search. With the Power My Analytics Bing Webmaster Tools connector, you can bring your Bing search clicks and impressions, your top queries and pages, ...

    • Power My Analytics MCP Server Guide

      The Power My Analytics MCP server lets you connect Claude Desktop, ChatGPT, and other Model Context Protocol (MCP) clients directly to your PMA Hub so you can ask questions about your marketing data in plain language. This guide walks you through ...

    • Amazon DSP Data Connector User Guide

      Amazon DSP (demand-side platform) is Amazon's programmatic advertising platform for buying display, video, audio, and streaming TV ads on and off Amazon. With the Power My Analytics Amazon DSP connector, you can bring your Amazon DSP orders, line ...

    • Mailchimp Data Connector User Guide

      Are you looking to streamline your email marketing analytics and reporting? Power My Analytics' Mailchimp integration allows you to easily connect your Mailchimp account, automate data collection, and create powerful reports. This guide will walk you ...