The search traffic decline that publishers spent 2024 arguing about is now settled fact, and the argument has moved on to what replaces it. Google still sends more visitors to the open web than every AI chatbot combined, by a wide margin. It just sends far fewer than it used to, and the drop has been steepest for exactly the kind of informational content most independent blogs are built on.
This piece lays out what the measured numbers show, why one site loses 60 percent while another loses 20, which responses are working, and which ones mostly waste a year.
What the search traffic decline actually looks like
The clearest picture comes from browsing data rather than opinion. The Pew Research Center tracked the real search behaviour of 900 US adults across roughly 69,000 queries and found that when an AI summary appeared, users clicked a traditional result on 8 percent of visits, against 15 percent when no summary was present. Clicks on the sources cited inside the summary itself came in at 1 percent. Sessions ended more often too: 26 percent of the time on pages with a summary, 16 percent without.
Publisher-side data tells the same story from the other end. The Reuters Institute’s 2026 trends report, based on a survey of 280 news leaders across 51 countries, recorded global Google search referrals to publishers down 33 percent year on year, with Google Discover down 21 percent. Chartbeat data through late 2025 broke the damage down by size, and the pattern is ugly for small operators: sites in the 1,000 to 10,000 daily pageview range lost around 60 percent of Google referrals, medium sites 47 percent, and large publishers 22 percent.
Two numbers are worth holding onto from that same research. Chatbot referrals, despite tripling over 2025, still amount to under 1 percent of publisher pageview referrals. And total traffic across the thousands of sites Chartbeat measures fell only 6 percent. Search collapsed. The web did not.
Why the numbers vary so wildly
A 22 percent drop and a 60 percent drop describe different businesses, and the gap is not random.
Query type explains most of it. Informational queries with clean factual answers are the ones AI summaries eat, because a 67-word answer genuinely resolves them. Transactional and navigational queries hold up better, since somebody comparing prices or looking for a specific tool still needs to land somewhere. A blog built on definitional posts is standing in the blast radius. A blog built on reviews, comparisons, tools, and community is standing to the side of it.
Site size matters through a different mechanism. Large publishers have brand search, apps, and direct visitors, so the share of their traffic exposed to search was smaller to begin with. A small site running on one channel takes the full hit.
Vertical matters as well. The Reuters Institute report noted that lifestyle and utility content, weather and TV listings and horoscopes, tends to have suffered more, which fits the pattern: content that answers a question in a sentence is content a summary can replace in a sentence.
There is a practical diagnostic in that. Take the twenty URLs that lost the most clicks over the past year and read the queries behind them. If those queries have short factual answers, the search traffic decline on that site is structural and the content mix has to change. If they are comparisons, product research, or questions where the reader wants a specific person’s judgment, something else went wrong and the fix is closer to ordinary SEO work.
What does not work
A few responses have been tried widely enough now to judge, and it saves time to name them.
Publishing more posts to make up the lost volume is the most common reflex and the least effective. If the format that lost traffic was the short informational answer, producing more of that format feeds the same machine. Volume without a reason also invites the spam policies aimed at scaled content.
Optimizing exclusively for AI citations is the second trap, less because it fails than because it pays badly. Getting cited is worth having, and cited brands do see some lift. But Pew measured citation clicks at 1 percent of visits. A citation is closer to a billboard than a doorway, and building a business on billboard impressions requires a different revenue model than the one most blogs run.
Waiting it out is the third. The publisher forecasting seen by Digiday in mid 2026 models AI Mode reaching around 80 percent of Google searches, with another 30 to 40 percent decline in Google referral traffic over the following few years. Forecasts miss, and that one may prove pessimistic. Planning as though traffic returns to 2023 levels is still a bad bet.
Where the traffic is actually going
The channels that hold up share one property: something other than an algorithm decides who sees the content.
Email is the strongest of them. A subscriber list is the only audience a blogger owns outright, and it converts better than search traffic ever did because the reader chose to be there. That choice comes with technical homework, since a list only works if the messages arrive, and email deliverability turns into the constraint the moment a list gets large enough to matter. The Chartbeat analysis flagged email, apps, and messaging as growing referral sources while search fell.
Social referrals stopped shrinking too. The Reuters Institute data showed them flat or slightly up globally, with X gaining 15 percent year on year and Facebook 9 percent, though Facebook is still far below where it stood in 2023. Video is the other pull, and publisher investment is concentrating on YouTube.
Direct traffic is the quiet one. It grows when people remember a site’s name, which is the slowest strategy available and the only one nobody can deprecate. It is also the channel most neglected during the years when search worked, because there was no pressure to build it. Sites that spent that period getting readers to remember who they were are the ones now describing the search traffic decline as survivable.
Discover deserves a caution rather than a recommendation. Some publishers report gains there, and the same Reuters Institute figures show it down 21 percent globally. It is a Google surface, so it carries a Google surface’s risk.
How to plan around the search traffic decline
None of this calls for panic, and most of it is ordinary work done in a different order. The one habit worth breaking is treating sessions as the number that matters, because a site can survive a search traffic decline of 40 percent with its revenue intact if the remaining visitors arrive through channels it controls.
- Check Search Console for which URLs lost impressions versus which lost clicks, since those are different problems with different fixes
- Sort the content library by query type and stop investing in the definitional posts an AI summary answers in a sentence
- Move effort toward formats a summary cannot replace: original testing, tools and calculators, opinion, data nobody else has, anything requiring a person to have done something
- Put an email capture on every page that earns attention, and treat list growth as the primary metric rather than sessions
- Build one non-search channel properly instead of four badly, choosing on where the audience already is
- Track referrals from AI assistants separately, because the volume is small but the intent tends to be strong
- Set a realistic target: a site that holds traffic flat through 2026 is outperforming its peer group
FAQ
How much of the search traffic decline is caused by AI?
Most of the measured drop lines up with the rollout of AI summaries and AI Mode, and both Pew’s click data and publisher referral data point the same way. Google disputes the framing, arguing that overall click volume to the web has stayed steady and that click quality has improved. Core updates and shifting user behaviour also play a part, so attributing everything to AI overstates a case that is already strong without it.
Will AI referrals replace lost search traffic?
Not on current numbers. Chatbot referrals grew sharply through 2025 and still sit below 1 percent of publisher pageview referrals, and one measure put Google’s search referrals at roughly 500 times the volume of ChatGPT’s. The visitors who do arrive from AI assistants tend to engage well, particularly on practical evergreen content, which makes the channel worth tracking without being worth betting on.
Should bloggers stop doing SEO?
No. Search still delivers more traffic than any other referral source, and the decline is a reduction rather than a shutdown. What changes is the target: ranking for questions with one-sentence answers has stopped paying, while queries where people need to compare, buy, use a tool, or trust a specific voice still send visitors.
How long does recovery from a search traffic decline take?
If the cause is a core update or a technical fault, months. If the cause is AI summaries absorbing the queries a site was built on, recovery in the old sense may not come at all, and the work is rebuilding around channels and formats that are not exposed to the same risk. Distinguishing between the two cases early saves a year of fixing the wrong thing.
