The clicks were real. The enquiries were not.

The complaint, and what is actually happening

The pattern turns up wherever immigration practices discuss paid search in public. Money goes out, clicks come in, and the people attached to those clicks are not people the practice can take on. The account gets cancelled, and the conclusion is that Google Ads does not work for immigration.

The experience is real. The conclusion is wrong. This is a matching failure, and it is visible in an account in about twenty minutes, for nothing.

Where to look

The campaign is set to a match type that lets Google decide what counts as related to your keywords, and Google decides generously.

You bid on something like immigration consultant. Google shows your ad to somebody typing how much is a work permit, can I apply myself, and cheapest way to get PR. Those are real people with real intent. It is just not intent you can convert, because they are shopping for a document and you sell representation.

The click is charged either way.

The screen that shows you this is the search terms report. In Google Ads, select Campaigns in the left navigation, then open Insights and reports, then Search terms. It lists the phrases people actually typed, not the keywords you bought. Read ninety days of it. In a badly matched account the gap between those two lists is obvious inside a minute.

One thing to avoid. Search terms insights is a different, summarised feature. You want the plain search terms report, the one with the query list.

That report is also where the fix lives. Every query describing somebody who wants to do it themselves is a negative keyword, and adding them costs nothing.

What changed in the account this month

Campaigns using automatically created assets or campaign-level broad match upgrade to AI Max for Search through September, and Google expects the eligible ones to finish by the end of the month. Those upgrades arrive with search term matching on by default, which widens the set of queries your ads can match.

For an account that already had a matching problem, wider matching makes it worse, and it does so without anybody pressing a button.

Worth saying plainly, because the opposite is being sold hard. We went looking for a clean measurement proving ad automation is what moved conversion rates this year, and it does not exist. The most-cited benchmark set reports the 2026 average Google Ads conversion rate as a level, 8.18 per cent, and it covers April 2025 to March 2026 across Google and Microsoft Ads, so it cannot settle a question about 2026 either way. That same source publishes attorneys and legal services separately at 5.55 per cent, which is the more useful row for this profession and the one worth measuring against.

Treat the upgrade as a change to decide deliberately, not as a proven villain.

It is probably not AI search

There is a popular explanation that AI answers are eating paid results. The best evidence available points the other way.

A pre-registered experiment on 1,065 United States participants earlier this year split them across a control group, a group with AI Overviews hidden, and an AI Mode group. When an AI Overview appeared, outbound organic clicks fell 39.8 per cent. Sponsored clicks did not move in any statistically meaningful way.

Its limits, so nobody sells it back to you oversold. A working paper, not peer reviewed. It measured browsing behaviour, not lead quality or signed agreements, on a United States panel. Best thing there is on the question, and not a study about your enquiries.

The line that moves on 5 October

Google posted an update to its Government documents and services policy on 4 August 2026 and begins enforcing it on 5 October. The update sets out who may advertise government documents and services directly.

The underlying restriction is older, and the standing policy page is the one that names the examples: Electronic Travel Authorization in Canada, passports, and residency lotteries.

An authorised provider may run those ads if it meets three tests. Its domain must be linked from an official, publicly accessible government website. It must promote only the specific document or service covered. And the ads must target only the matching region. The policy states explicitly that a commercial contract, a business or trade licence, and an entry on a business registry are not valid authorisation, whether or not the registry is government-managed.

Read that last part twice before anyone tells you a College listing settles it. The policy does name official regulatory portals identifying government-sanctioned specialists as a valid category, so read it against your own situation rather than taking a blanket answer from an agency, this one included.

In practice the exposure sits in the words in the ad rather than in your account settings. The operative distinction is representation and advice against obtaining a document. Copy offering professional representation, an assessment, or advice on a stream sits on one side. Copy offering to get somebody a named government document sits on the other. That distinction is also what fixes the matching problem.

One thing that is switched on and nobody told you

Ads inside Google AI Overviews reached Canada. Google's help page states they are "currently available in English on mobile and desktop devices in the following countries", and Canada is one of the twelve named. There is no opt-in, and text and Shopping ads from existing Search, Shopping and Performance Max campaigns are eligible automatically.

Eligible is not the same as showing. Ads appear there only on a subset of queries, and Google's excluded-vertical list ends in the words "and more", so where immigration sits is unconfirmed rather than confirmed. A French-language practice is not eligible at all, because the placement is English only for now.

The part worth knowing. Google gives you no segmented reporting for it. Ads in AI Overviews are reported as Top Ads, so that line in your account already includes them and always has.

The order to do this in

None of these costs money.

Read ninety days of search terms. Campaigns, then Insights and reports, then Search terms. Compare what people typed against what you bought.

Add negatives for do-it-yourself and document-shopping queries.

Check whether any campaign runs automatically created assets or campaign-level broad match. Those upgrade this month. Decide it deliberately and keep a clean before-and-after period.

Read every headline against the acquisition-of-a-document line, before 5 October.

Then check the destination. Ads send traffic to a page. If the page does not say what you do, for whom, and on what authority, paid traffic buys more people the chance to leave. In our July 2026 sweep of 195 licensed RCIC websites, 150 carried no R number anywhere we could reach. That sample came from our own outreach research rather than a random draw, so read it as a floor rather than a rate.

When the answer is to stop

Sometimes it is, and stopping is not a failure. Paid search suits a practice that already converts the traffic it has. If the page is not converting today's visitors, buying more is an expensive way to find that out. Fix the page, then buy the traffic.