- Definitional pages held impressions and lost clicks, which is the signature of an answer given before the click.
- Queries containing a supplier decision have barely changed in behaviour.
- Generic content became free to produce and therefore worth almost nothing; two hundred fine articles can still rank for nothing.
- Answer-first structure, real tables and specific figures are what makes a passage quotable.
- Publishing more is the wrong response to falling content traffic; our own site is smaller than in 2022 and produces more enquiries.
Two years into AI answers being the default for a large share of queries, enough has settled to say something useful. This is what we have seen in our own accounts and on our own site, rather than what the industry press says.
What actually fell
Informational, definitional queries. Across the client sites we manage, articles answering what is X and how does Y work have lost a substantial share of their clicks while impressions held or rose. That pattern, impressions steady and clicks down, is the signature of an answer being given before the click.
It is not a catastrophe. It is a re-pricing. Those articles were always the least commercially valuable traffic we brought in. They just used to look impressive in a report.
What did not fall
Queries with a supplier decision inside them. Shopify migration cost, react developers noida, how much does an ecommerce site cost, anything with near me attached. Behaviour on these has barely changed. Someone spending money still wants to see who they would be spending it with, compare two or three, and read something specific enough to trust.
The same holds for anything requiring current, local or first-hand information. A model can summarise what an ecommerce migration involves. It cannot tell you what it costs from a studio in Noida this quarter, or what went wrong on the last one.
Cheap content became worthless content
The more consequential change is on the supply side. Producing a competent, generic 1,200-word article now costs almost nothing, which means a competent, generic 1,200-word article is worth almost nothing. There is unlimited supply and no reason to prefer yours.
We see the result in audits: sites that have published two hundred articles in eighteen months, all technically fine, none ranking, none linked to by anyone, and which collectively make the site look thinner rather than deeper.
The conclusion follows directly. If a piece of content could have been produced without knowing your business, it is not an asset.
What we publish now
Things only we can say. Numbers from our own projects. What a decision cost when we got it wrong. Ranges we would actually quote. That is the whole of the advantage.
Decision content over explanation content. Comparisons with a recommendation attached. Cost breakdowns. Lists of questions to ask a supplier, including the awkward ones about us.
Updates to what already exists. A page with two years of history and a handful of links responds better to genuine improvement than a new URL does to being published.
Original data, once a year. A benchmark across our own client base, published with the method attached. It is the only format that reliably earns links, and links are the thing AI has not made cheap.
Structure for being quoted
If part of your audience now meets your content as a sentence inside a summary, being extractable matters.
Answer the question in the first forty words under the heading. Use real tables for comparisons rather than prose lists. Give specific figures where you can: a model can quote four to eight weeks and can do nothing with it depends. Keep sections self-contained. Say when the page was updated and what changed.
We restructured six client articles this way and tracked them across four months. Four picked up citations in AI answers. Clicks to those pages were slightly down, enquiries mentioning them were up. One small sample, not a law, and the structural work costs an hour a page.
We restructure existing content for answer engines and write the decision-stage pieces that still earn enquiries, with the numbers your team actually has.
Volume is the wrong lever
The most common reaction we see to falling content traffic is to publish more. It is precisely wrong. More generic pages dilute the site, consume the internal link budget, and add maintenance debt.
Our own cadence is one substantial article a month with occasional short notes, plus a scheduled update cycle. The site is smaller than it was in 2022 and produces more enquiries. The sizing method behind that is in a content strategy that survives quarter one.
What this means for a small team
If you publish once a month with one person writing, the change is smaller than the industry noise suggests. You were never going to win on volume, and volume is what has been devalued. What has become more valuable is exactly what a small team has and a content farm does not: a named person with opinions, client work to describe, and prices they will state in public.
The practical adjustment for a one-article-a-month operation is to move about half the output from explanation to decision content, add real figures wherever you have them, and spend the time saved updating the three or four pages that already earn something.
What we do not know yet
Whether citation in an AI answer produces measurable business value at scale. Whether current click patterns are stable or transitional. Whether the models will keep surfacing sources as prominently as they do now.
Anyone claiming certainty there is guessing with more confidence than the data supports. What we are confident about is narrower: first-hand experience, specific numbers and a clearly identified business are the inputs that have become more valuable, not less.
If you want to act on this
Start with an inventory rather than a plan. Sort your pages by clicks over the last year, mark which are definitional and which carry a supplier decision, and be honest about which pile most of your effort has gone into. Most of the sites we audit are eighty per cent pile one.
Moving that ratio is the whole project. It is what our AI search optimisation work does on existing content, and the structural half of it is covered in advanced SEO.
Questions
Has AI killed blogging for business?
No, but it has killed generic blogging. Content that could have been written without knowing your business now competes with unlimited free supply. Content containing your own numbers and experience competes with nothing.
Should I block AI crawlers from my site?
For most businesses, no. Being absent from the answers sends nobody to your site. It removes you from consideration. Blocking makes sense mainly where the content itself is the product being sold.
How do I get cited in AI answers?
Answer directly under a clear heading, use specific figures, put comparisons in tables, keep sections self-contained, and run a site with an unambiguous identity. There is no submission process. It is structural.
Is long-form content still worth writing?
Length is not the variable. A 600-word piece with real numbers beats a 3,000-word summary of common knowledge. Write to the length the question deserves.