Build your report backwards from the booking

The mid-year report is due and the middle of it no longer explains itself. The demand at one end and the booked work at the other are as visible as they ever were. The stretch in between, from the question somebody asks to the page they land on, went quiet in the last month.

AI Mode came out of Labs at I/O on 20 May and is now in front of everyone searching. Google has since said that clicks from AI Mode are counted in Search Console the same way as other web results, folded into the totals you already look at. At Marketing Live the following day, AI Max arrived in Search ads, which loosens matching further in the name of finding queries you did not think of.

Between them they take the middle of the path out of view and leave the two ends alone. So build the report from the booking backwards, and stop asking the middle to account for itself.

What the Search Console change gives you, and what it does not

It gives you honesty in the totals. AI Mode traffic is not missing from your numbers, and impressions there are not being hidden.

It does not give you a breakdown. There is no AI Mode filter, no separate report, no way to say how many of this month’s clicks came from a conversational answer rather than a blue link. Anyone offering you that number is estimating.

So change what you use Search Console for. It is a demand and trend instrument now, not an attribution source. Three things on it are still worth reading every month:

• Impressions against clicks. Impressions rising while clicks stay flat is the normal shape this year, not a fault. It tells you that you are being seen inside answers you do not get a visit from.
• Query length and shape. Longer, more specific questions showing up is the surface changing. It is also a content brief: those are the questions your site should answer plainly.
• Branded impressions and clicks. People who get recommended in an answer and then look you up by name arrive as brand searches. Brand volume trending up while non-brand clicks drift down is a good month, not a bad one, and it is invisible if you only look at the site total.

Build the report backwards from the booking

The report we send has six inputs. Only the last two involve a platform.

1. Fix the unit. A booked appointment or a booked job, not a click, a lead or a form fill. Everyone in the office should be able to name the number for last week.
2. Capture source at the moment of contact. A tracked number with dynamic insertion on the website, a hidden source field on every form, and a “how did you hear about us” question asked by whoever answers the phone. Three imperfect signals beat one perfect one you do not have.
3. Keep the click id. Store the Google click id with the lead when the form comes in. It sits in the record doing nothing until step 5, then it becomes the only clean line between spend and a booking.
4. Write the outcome back to the lead. The booking, the show, the treatment or job type, the value band. This is CRM discipline, not analytics, and it is where most accounts fall apart. If the front desk does not update the record, there is no report to build.
5. Send the outcome back to the ad platforms. Offline conversion import for the bookings that started with a click. Automated bidding is only as good as the signal it gets, and a platform told about booked work bids differently than one told about form submissions.
6. Keep an unknown bucket and watch its size. Some bookings will never be attributable. Report them as their own line rather than dividing them up by guesswork. The bucket growing slowly is expected. The bucket doubling in a month means something broke in steps 2 or 4.

The one page we send every month

No dashboards. One page, same shape every month, so the trend is readable at a glance:

• Booked jobs this month, last month, same month last year.
• Booked jobs by source, with the unknown bucket shown honestly.
• Cost per booked job by paid channel, not cost per lead.
• Calls answered against calls missed, by hour of day.
• Search Console: impressions, clicks, brand against non-brand.
• Profile actions: calls, direction requests, website clicks.
• The one thing we changed last month and what happened after it.

The last line is the one owners read first. It is also the reason we only make one change at a time.

Reading it when the numbers disagree

They will disagree. The platform will claim more conversions than you have bookings, because it counts a form and you count a patient in a chair. That gap is not fraud, it is two different questions.

When you genuinely cannot tell whether a channel is working, stop arguing with the attribution and run a test instead:

• Pause one channel for two weeks and watch total booked work, not that channel’s own numbers.
• Or split by area: run the channel in half your service area and not the other half, then compare booked work per area.
• Do it in a quiet two weeks, not in your busiest month, and never during a core update.

That is a blunt instrument, and it is more reliable right now than any model of a path that ends inside a written answer.

What to do this month

• Write down your half-year numbers using booked work as the unit, before anyone asks for a report.
• Check that your forms store the click id and a source field, and that both are still arriving.
• Add “how did you hear about us” to the phone script and log the answer in the same place as everything else.
• Turn on offline conversion import for at least your main search campaign.
• Split your Search Console view into brand and non-brand, and record where both stand today.
• Pick the one channel you argue about most and schedule a two-week pause test for a quiet stretch.

The reporting work is the only thing keeping Google Ads honest while the click path gets murkier. It depends on the CRM being kept up to date, which is why most of our measurement work is really CRM work, and on the business being described accurately enough to get recommended in the first place, which is the other half of visibility in AI answers.