The most common question we get from marketing leaders right now is whether SEO still works. It is a fair question. AI summaries answer a growing share of queries without a click, and a meaningful number of people start research inside an assistant rather than a search box.
The honest answer is that a specific type of SEO stopped working, the rest matters more than before, and the strategies most damaged were the ones that were always weakest.
What genuinely got worse
Content that answered a simple factual question is largely finished as a traffic strategy. If a query has a short, unambiguous answer, an AI summary now provides it and the click does not happen. Sites built on definition posts, conversion calculators and basic how-to content have seen real declines that are not coming back.
Broad top-of-funnel content is under similar pressure. The 'ultimate guide to X' aimed at people with no purchase intent was always a weak commercial play; it is now also a weak traffic play.
What matters more than it did
- Commercial-intent queries. Someone comparing vendors or checking pricing still clicks through, because the decision requires more than a summary paragraph.
- Genuine expertise. AI systems synthesize from indexed sources. Being one of those sources requires having said something the model could not have assembled from generic material.
- Original data. Proprietary research, benchmarks and survey results get cited because there is nowhere else to get them.
- Brand and entity strength. When several sources say similar things, recognizable and well-referenced entities get selected disproportionately.
- Structure. Clear headings, direct answers, tables and specific claims are more extractable — and extractability is now a ranking-adjacent property.
Being cited is the new being ranked
When an assistant answers a question about your category and names three vendors, being one of them is worth more than a blue link ever was — it arrives with an implicit recommendation. That placement is influenced by a mix of factors, most of which are the same things that have always constituted good SEO plus a few that are newer.
- Be present on the sources these systems draw from most heavily — including review platforms, industry publications and community discussion, not only your own site.
- Make claims specific and verifiable. Vague marketing language does not survive synthesis; concrete facts do.
- Keep entity information consistent — company name, description, leadership, locations and offerings — across your site, structured data and third-party profiles.
- Publish things worth quoting: original numbers, clear frameworks, direct positions on debated questions.
Measurement has to change too
Impressions rising while clicks fall is now a normal and not necessarily bad pattern — it can mean your content is being summarized, which still builds familiarity even without the visit. Judging the program on sessions alone will produce the wrong conclusion.
We track branded search volume, direct traffic, assisted conversions and pipeline alongside organic sessions. When summaries reduce clicks but organic pipeline holds or grows, the program is working and the traffic metric is simply no longer measuring what it used to.
What we would do with a fresh budget today
- Fix technical foundations first — nothing works if the content cannot be crawled and rendered.
- Publish bottom-funnel comparison and evaluation content before anything else. It converts fastest and is least affected by summarization.
- Invest in one piece of original research per year. It is the most reliable way to be cited by both journalists and AI systems.
- Build genuine expertise into content through subject-matter interviews rather than research aggregation.
- Treat review platforms and industry directories as part of the SEO surface, not a separate reputation task.
The strategies AI search damaged were mostly strategies that produced traffic without producing customers.
For companies whose organic program was already built on genuine expertise and commercial intent, remarkably little has changed. For programs built on volume and optimization, quite a lot has.