3 Predictions For 2027, And Why You Won’t Be Able To Check Them via @sejournal, @DuaneForrester

Every practitioner reading this has already placed a bet on which interface survives. You made it when you decided where the effort goes this quarter, whether AI answers are a channel or a curiosity, and how much of your reporting still quietly assumes a click. Google has placed the same bet, with considerably more money on the table.

What follows is tea-leaf reading rather than certainty. Three predictions about where Google sits by the end of 2027, and a fourth about whether anyone will be able to tell if the first three were right. I build measurement tooling in this space (CitationIQ), so weigh the back half of this accordingly, as I do have a vested interest.

The thing that sent me back to this question was a recent article from Greg Jarboe about winners and losers in U.S. website traffic. I have reservations about how some of the traffic data in it was assembled, and I am not going to lean on those figures. But it raises something worth taking seriously, which is what happens to attribution when the surfaces sending people to you stop saying so.

The Money Is Not Moving Where The Traffic Moved

Alphabet reported Q2 2026 on July 22. Google Search and other advertising revenue came in at $63.3 billion, up 17% year over year, a full year into AI Overviews and the global rollout of AI Mode. Whatever the cannibalization thesis predicted, this is not it.

Read the second line, though. That 17% is the first deceleration in six quarters, and the CFO flagged a harder comparison coming in Q3. This is not a collapse, but it is the first visible bend in a curve that has been close to straight for years.

My prediction for 2027 is that both things stay true at once. Search advertising revenue keeps growing in absolute terms, and the share of it earned by sending a person to somebody else’s website keeps shrinking. Google gets better at monetizing intent at the moment it is expressed rather than at the moment it is handed off. The revenue does not migrate away from Google. It migrates away from the part of the arrangement that ever paid you.

What would prove me wrong: two consecutive quarters of negative Search revenue growth with no obvious macroeconomic cause. That would suggest the answer layer is eating the ad business rather than absorbing it.

Ranking And Earning Stopped Being The Same Job

The old arrangement had a clean shape. Google ranked you, the ranking sent traffic, and Google sold ads against the intent that traffic represented. Every party in that arrangement needed the click, which is precisely why the click was measurable. It was the unit of exchange, so everyone had a reason to count it.

The answer layer breaks the arrangement without breaking the revenue. Intent gets monetized where it is expressed. Most searches now end without a click-through to an independent site, and the advertising business grew anyway. Those two facts sitting next to each other are the entire story.

So the stack decouples. The surface that ranks and the surface that earns are no longer the same surface, and they are no longer under the same pressure to behave consistently with one another. By 2027 I expect the ranked results page to persist largely as a legacy interface, maintained because advertisers and two decades of habit require it, while the consequential product decisions happen somewhere else entirely.

What would prove me wrong: referral volume to independent publishers rising alongside advertising revenue. If both climb together, the old arrangement is intact and I have misread the mechanism.

2 Companies Are Betting Against Each Other On You

Google cannot walk away from its own stack. The index, the crawl, the auction, the advertiser relationships, the habits of a generation of users, all of it is an asset and an anchor at the same time. The companies building answer engines will not rebuild traditional search underneath their products either, because that would mean constructing the thing they exist to replace.

Neither side can hedge. Each is betting that the average person lands on their side of the split, and both bets get settled by the same population making the same small daily decisions about where to type a question.

My prediction is that neither side wins outright by 2027 and both continue operating as though they will. Most organizations end up running parallel: optimizing for a ranked surface and an answer surface simultaneously, with different assumptions and a shared budget. That is not a transitional state on the way to a flip. For most companies it is the steady state, and planning for it as though it were temporary is its own kind of error.

What would prove me wrong: a decisive consolidation of query volume onto one interface type, visible in independent panel work rather than vendor claims.

Every Prediction Above Has The Same Problem

Someone is going to ask you next year whether any of this actually happened. You will reach for a number. That is the right instinct, and it is exactly where the trouble starts, because the instruments most of us reach for are being replaced while we reach for them, and the replacements are not being announced.

So here is the fourth prediction, and the one I hold with the most confidence. By the end of 2027, this industry will be making more of its decisions on inferred data than at any point in the last twenty years, and reporting those decisions with the same confidence it used to reserve for data it actually counted.

A Counted Number And An Inferred Number Are Not The Same Kind Of Evidence

A counted number comes from an event that was observed on infrastructure you or your vendor control. An inferred number comes from a sample extrapolated to a population nobody can fully enumerate. Both can be correct. Only one of them can be checked.

Most people treat this as a quality ranking, where counted is good and inferred is suspect. It is not that. Panel work answers questions server logs cannot touch. The distinction matters for a narrower reason: it determines what you can do when a number surprises you. With a counted number you can go and find the underlying events. With an inferred number, you can ask the vendor about their method and hope one has been published.

Google Analytics is the useful example here precisely because it contains both classes and displays them identically. Session counts are close to counted. Channel attribution is inferred, and Direct is the bucket where that inference fails quietly rather than loudly. Nothing in the interface tells you which column you are looking at. Same font, same chart, same export, same two decimal places.

Then look at what happened to that tool this year.

On May 13, Google added an AI Assistant channel to the default channel group. Useful change, long requested. It also arrived with the full list of recognized AI referrers unpublished, no statement about how that list would be maintained, and the channel definitions page not yet updated to describe it. Within weeks, the set of platforms being reported differed from the set named at launch. The channel counts forward only, with no backfill of history.

That is a counted number whose definition moved without anybody being told, inside a tool that reports the result to two decimal places.

Then on September 1, standard reports showed zero traffic across a very large number of properties. Realtime kept showing users the whole time. Collection was working. Reporting was not. Those are separate systems, and they fail separately, and a practitioner who understands that can explain the zero to an executive in a single sentence. A practitioner who does not spends a day re-checking tags that were never broken. A week later, there was still no public confirmation that the affected data would be restored or backfilled.

Neither of those is a scandal. Both are ordinary. That is what makes them worth learning from.

The Claim Nobody Went And Checked

When Google launched AI Mode in May 2025, its citation links carried a noreferrer attribute, which stripped the referrer and dropped those clicks into Direct. Practitioners caught it quickly. John Mueller said publicly that it looked like a bug on Google’s side. The attribute came off within days, and named practitioners confirmed traffic classifying as organic again.

Through 2026, a substantial volume of vendor content has asserted that AI Mode strips attribution by architectural design, that the choice is deliberate rather than accidental, and that no workaround exists. None of it references the correction. A claim that was accurate for roughly one week has been inherited forward for a year and hardened into permanent design intent along the way.

Whether the attribute sits on those links today is beside the point. The point is that almost nobody repeating the claim knows either way, because settling it means opening the page and reading the markup, which takes about ninety seconds. They are not reporting an observation. They are repeating one.

What I Am Not Going To Give You

A test. I worked through what one would look like, and the honest answer is that I don’t think a definitive version can be built. There is no ground truth to score against, which means a test you run can never fail. Provenance is frequently undisclosed, so any step that says identify the population fails at the vendor’s discretion rather than your competence. Most real numbers are hybrids, counted and then modeled inside the same figure. And any classification you make expires the moment a vendor changes method, usually in release notes nobody reads.

Publishing a checklist here would reproduce the exact error the rest of this piece is describing.

What does hold up is a set of questions you put to the vendor, or to the account, before a number goes into a deck. What population does this describe, and can it be named? Was the value observed or extrapolated? What would move this number if the world outside stayed exactly the same? And when the system cannot determine an answer, what happens to that session: does it drop out of the dataset, or get filed somewhere quiet where someone will later read it as a finding?

That last question is where Direct traffic lives, and it is why Direct has spent the past two years carrying an argument it was never built to support.

None of these produce a score. They produce either an answer or a silence, and the silence is itself the finding. Someone who applies them badly still ends up asking a vendor something the vendor ought to be able to answer, which is a lower bar than a test and a considerably more durable one.

The three predictions above are worth arguing with. What is not worth arguing about is that by 2027 you will be defending or abandoning them using numbers you did not observe, produced by methods you cannot inspect, presented in interfaces that will not tell you which is which. Learning to tell those apart is not an analytics skill anymore. It is most of the job.

More Resources:

This post was originally published on Duane Forrester Decodes.

Featured Image: Andrey_Popov/Shutterstock; Paulo Bobita/Search Engine Journal