You can export a Google Ads campaign report, paste it into ChatGPT or Claude and get a tidy summary. The problem is that the AI will speak with the same confidence about what is clearly visible in the numbers and about what it is only guessing. So a useful analysis needs three things: the right columns, context that is not in the table, and a prompt that makes it label every statement as an observation, a hypothesis or missing data.
Below: what to export, what to write next to the table, a copy-paste prompt, an illustrative example, and how to check whether the AI calculated correctly. If you need to review not campaigns but the specific searches that triggered your ads, there is a separate article on search terms and negative keywords (Google Ads search terms and negative keywords).
What to export from Google Ads and what the columns mean
In the campaigns or ad groups table, choose a date range, add columns and click the download button. Google Ads lets you download in CSV, Excel (XLSX) or Google Sheets format (checked October 22, 2026). Campaign level is enough for a first analysis; ad group level is useful once you know which campaign you want to dig into.
- 1Campaigns
- 2Download icon above the table (right)
- 3Download
- 4CSV, XLSX or Google Sheets
Based on the Google Ads Help Center (checked October 23, 2026). Choose the date range and columns before downloading. The "Schedule" option in the same place lets you receive the report by email on a regular basis.
Below are the columns you need and how the Google Ads Help Center defines them. It is worth pasting these definitions into the prompt too, so the AI does not mix them up.
| Column | What it means |
|---|---|
| Clicks, impressions | How many times the ad was shown and clicked |
| CTR | Clicks ÷ impressions |
| Cost | How much was spent in the period |
| Conversions | Number of primary conversion actions |
| Conv. value | Sum of the conversion values in the Conversions column |
| Cost / conv. | Average cost of one conversion: cost ÷ conversions |
A conversion is an action you count as a result (an enquiry, a call, a purchase). Important: the "Conversions" column contains only primary actions, while "All conv." also includes secondary ones. If you paste both columns into the AI without explanation, it may mix them up. CPA (cost per action) is the same as "Cost / conv.".
Impression share is the share of possible impressions your ads received. It is worth adding if you are asking whether the budget limits the campaign. Google says data for these metrics may take 24–48 hours to update, and with low traffic you may see a dash instead of a value.
What context does the AI need that is not in the table?
Most wrong AI conclusions come not from a bad model but from the fact that it does not know what happened in the account. Before the analysis, write down answers to these questions and paste them in together with the table.
- The date range and changes during it. When did you change the budget, bid strategy, ads or landing page? A campaign whose budget you doubled mid-month cannot be compared as if it were an even month.
- What counts as a conversion. Which actions are primary: form submission, call, purchase? Did you change conversion settings during the period? If measurement started or changed midway, the numbers before and after are not comparable.
- Attribution model. This is the rule for how a conversion is split between clicks. Google says the default for most conversion actions is the data-driven model, and the other option is last click. First click, linear and other older models are no longer supported. With the data-driven model the number of conversions can be fractional, and the AI needs to know this so it does not treat 3.4 conversions as an error.
- Conversion delay. Google Ads assigns a conversion to the day of the ad impression or click, not the day the customer actually sent the enquiry. So the numbers for the last few days are still incomplete: Google says delayed conversions make CPA look higher and return look lower than they really are. You can see how long your customers take with the segment "Conversions → Days to conversion", choosing a date range that ended at least 30 days ago.
- Seasonality and outside events. Is this a normal month for your business? A sale, holidays, the start of the school year, a competitor's campaign: the AI does not know about these until you tell it.
A prompt that separates facts from guesses
This prompt requires one of three labels on every statement and a separate assessment of whether there is enough data. Replace the square brackets with your own information.
You are a careful Google Ads analyst. Below are a campaign report and context. Context: date range [dates]; changes during the period: [budgets, strategies, ads, with dates]; primary conversions: [which actions]; attribution model: [data-driven / last click]; typical time to conversion: [if you know]; seasonality: [what is typical for this period]. Definitions: CTR = clicks / impressions. Cost / conv. = cost / conversions. Conversions = primary actions only. Task: 1. Write a summary of up to 8 points. Label each point: OBSERVATION (directly visible in the data, with numbers), HYPOTHESIS (a possible explanation that needs checking; say how) or MISSING DATA (what is needed to decide). 2. For each campaign, assess whether there are enough conversions and clicks to draw conclusions, and say why. 3. Give every calculated number with its formula, e.g. 512.00 / 16 = 32.00. 4. Do not suggest pausing campaigns or ads based on CTR or a low number of clicks alone. 5. Do not invent market averages or benchmarks. If you do not know something, write MISSING DATA. Data: [paste the table]
Point three matters for checking: when the AI has to write the formula, you see a mistake straight away. Points four and five guard against two of the AI's most common habits: bold recommendations from a few clicks and made-up "average CTRs in your industry". More on prompt structure in the guide on how to write prompts (How to write AI prompts).
- 1Download the campaigns tableGoogle Ads → CSV / XLSX
- 2Add contextChanges, conversions, attribution, seasonality
- 3Prompt with three labelsChatGPT / ClaudeObservation, hypothesis, missing data
- 4Recalculate the numbersGoogle Sheets / Excel
- 5A person makes the decisionBased on your own sales data
Illustrative example: input and answer
Say a language school in Klaipėda (a port city in western Lithuania) advertised courses with four search campaigns in September. The conversion is the registration form. The budget did not change during the month. The numbers are made up.
| Campaign | Clicks | CTR | Cost | Conv. |
|---|---|---|---|---|
| English courses | 640 | 4.1% | €512.00 | 16 |
| German courses | 210 | 3.2% | €189.00 | 3 |
| Brand | 150 | 18.5% | €36.00 | 9 |
| Kids' camp | 38 | 1.2% | €41.80 | 0 |
With the prompt above, the answer could look like this (shortened):
OBSERVATION. The English campaign accounts for the largest share of spend (512.00 / 778.80 = 65.7%) and brought 16 of the 28 conversions. Cost / conv. 512.00 / 16 = €32.00.
OBSERVATION. The brand campaign has the lowest cost / conv.: 36.00 / 9 = €4.00.
HYPOTHESIS. Some brand searches may have brought people who already knew the school and would have come without the ad. A simple comparison of periods with and without the campaign is only a hint, not proof of a causal effect: season, demand and other campaigns change between periods. A more reliable answer would need a controlled experiment planned in advance.
OBSERVATION. The German campaign's cost / conv. is 189.00 / 3 = €63.00, almost double the English one (€32.00).
Sample uncertainty. This difference is calculated correctly, but it rests on only 3 conversions. With such a small number, one registration more or less changes the result a lot: with 4 registrations the cost would be 189.00 / 4 = €47.25, with 2 it would be €94.50. So the number describes last month, but does not yet show that the German campaign is fundamentally more expensive.
MISSING DATA. Kids' camp: 38 clicks and 0 conversions are not enough for a conclusion. A low CTR (1.2%) does not by itself show the campaign is bad. You need to know when people usually register for camps and look at which searches triggered the ad.
An answer like this is useful because it separates what is visible in the numbers from what needs more data or an experiment. But it does not make decisions for you. For example, the English campaign collected the most registrations, but that alone is not a reason to raise its budget: the report does not tell us how many registrations turn into students or what cost per registration is acceptable to you. You answer those questions, using your own sales data.
How to check the numbers the AI gives you
Language models do not calculate like a spreadsheet: when adding up longer columns, dividing or working out percentages they sometimes get it wrong, and the error looks convincing. For example, the AI could write the overall cost / conv. as €28.90, when the correct figure is 778.80 / 28 = €27.81.
- Paste the same table into Google Sheets or Excel.
- Recalculate every number the AI mentions with a formula, e.g. for cost / conv. =D2/E2, for the overall figure =SUM(D2:D5)/SUM(E2:E5).
- Compare with the AI's answer and with the Google Ads interface. If they do not match, correct it in the AI chat and ask it to redo the conclusions with the right numbers.
- Check that the AI did not mix up columns ("Conversions" and "All conv.") and did not add campaigns that are not in the table.
If you do this report regularly, it is worth doing the calculations themselves in a spreadsheet and giving the AI an already calculated table, asking only for interpretation. That way it does what it is better at: writing and structuring.
Why not pause ads based on CTR or a few clicks
Google itself says that a good CTR depends on what you advertise and on which network. A low CTR can mean the ad shows for broad searches, but the people who click do register. A high CTR can mean curious visitors who will never become customers. Conversions and their cost matter more for a decision than CTR.
With small numbers, one conversion changes the picture completely, as you saw in the German example. There is no universal threshold for how many conversions make a conclusion reliable: it depends on your price, sales cycle and traffic. In practice: if a campaign has a handful of conversions, extend the date range or wait first instead of pausing. And before judging a "zero", check that conversion tracking really works.
How to compare date ranges correctly
The request "compare September with August" looks simple to the AI, but the result can mislead. Before comparing, check:
- Same length and days of the week. 30 days against 31, four weekends against five: that is already a difference. It is better to compare periods of the same length that start on the same day of the week.
- The last days without delay. Because conversions are assigned to the ad day, not the conversion day, a period that has just ended will show fewer conversions than will eventually build up. Either wait, or do not compare the last few days.
- Seasonality. If the business is seasonal, it makes more sense to compare with the same period last year than with last month.
- Changes between periods. Changes to budget, strategy or conversion settings must be written down in the context, otherwise the AI will explain the difference with something else.
- The same conversion timing logic. If you compare results with Google Analytics, know that it counts conversions by event date. Google Ads offers "Conversions (by conv. time)" columns, which line up better with other tools.
Compare these two periods of equal length. For each campaign, give the change in absolute numbers and as a percentage, with the formula. List separately the changes that could be explained by the changes I listed or by seasonality, and those with no explanation in the data. Mark campaigns with fewer than [your chosen number] conversions in both periods as an insufficient sample. Changes between periods: [list with dates]. Period A: [table] Period B: [table]
Data you should not paste, and when automation is worth it
A campaign report usually has no personal data, but still do not paste your account number, logins, customer lists or CRM exports with names and contacts into outside AI tools. If campaign names contain client or partner names, replace them with neutral ones.
If you do this review every week or month, you can cut down the manual export. Google Ads itself lets you schedule a statistics table to be sent by email (for example daily or weekly, in CSV or PDF format). The next step is a report in Google Sheets or Looker Studio that fills in automatically; we wrote about such tools separately (Business report automation tools). And finally, a technical integration through the Google Ads API or n8n that collects the data and prepares it for analysis. But the "observation / hypothesis / missing data" logic stays the same: a person makes the decision on the budget or on pausing a campaign.
Sources
Google Ads Help: Understand your conversion tracking data: definitions of conversions, conversion value, cost / conv., conversion rate and all conversions.
Google Ads Help: Clickthrough rate (CTR): Definition: the CTR formula and the fact that a good CTR depends on what you advertise and the network.
Google Ads Help: Impression share: definition, data updates within 24–48 hours.
Google Ads Help: Primary and secondary conversion actions: what goes into the "Conversions" column and what into "All conv.".
Google Ads Help: About attribution models: data-driven and last click models, models no longer supported.
Google Ads Help: Conversion delay estimates: conversions by impression date, the effect of delay on CPA and return.
Google Ads Help: How long customers take to convert: the "Days to conversion" segment, a date range that ended at least 30 days ago.
Google Ads Help: Data discrepancies: conversions assigned to the ad date rather than the conversion date, the "by conv. time" columns.
Google Ads Help: Create, save and schedule reports: download formats and scheduled email reports.