Ranking first on Google used to be the whole game for a software company. You published pages, Google ranked them, buyers clicked, and some of them signed up. That pipeline is leaking. The clicks are going down fast, and buyers have already moved: they ask ChatGPT what to use. This note walks through the numbers, explains why getting named inside an AI answer is becoming the entry point that matters, and lists what a developer tools company can do about it.

What happened to search clicks

Google increasingly answers the question on the results page itself. When it shows an AI summary, the top organic result loses most of its clicks. Ahrefs measured a 58% drop for the number-one position in its December 2025 data, published in February 2026. The same team, using the same method, had measured a 34.5% drop a year earlier. The fall is getting bigger, not smaller.

Then there are the searches that never produce a click at all. SparkToro tracks this with clickstream panels. In the first months of 2026, 68% of Google searches ended with no click on any result. In 2019 it was 49%. SparkToro itself points out that these numbers come from different panels across the years, so compare the direction, not the decimals.

Pew Research Center studied real browsing histories instead of search logs. People who saw an AI summary clicked a normal result 8% of the time. People who did not see one clicked 15% of the time. Only 1% clicked the sources listed inside the AI summary. Read that last number again if you ever hear someone say "we rank, so we get the traffic."

Buyers changed where they start

The click numbers describe search. The buyers describe something bigger. G2 surveyed 1,076 B2B software buyers in April 2026. 51% said they now start software research with an AI chatbot more often than with Google. A year earlier, that number was 29%.

6sense asks buyers a different question: do you use AI tools at any point while buying? 94% of B2B buyers said yes in 2025. The two numbers measure different things. The first is about where research starts, the second about whether AI shows up anywhere in the process. Both moved the same direction.

Developers are no exception. Stack Overflow's 2025 survey covered 49,000 developers in 177 countries. 84% use or plan to use AI tools. 46% say they don't trust the accuracy of AI output, and 66% report getting answers that are "almost right, but not quite". Both habits matter here: developers ask AI constantly, and then they verify by hand. Each habit leaves traces a company can show up in.

Two words worth separating: named and cited

When ChatGPT answers "what's a good tool for feature flags", two different things can happen. The answer text can name your product. The answer can also list your page as a source it used. People in this field call the first a mention and the second a citation. A mention without a citation usually comes from the model's memory of its training data. A citation means the model read your page while answering. They break for different reasons, and they get fixed differently, so track them as two columns in your notes.

Why getting named beats ranking now

If buyers start in AI, the products named in the answer are the ones they go look at. G2's report says it directly: vendors that never appear in AI answers often fail to make the candidate list at all.

There is click data behind this. Seer Interactive tracked 3,119 queries across 42 client sites for 15 months. Brands whose pages were cited in Google's AI summaries lost 49.4% of their organic clicks year over year. Brands that were not cited lost 65.2%. Being cited did not make the problem disappear, but it cut the loss by about a quarter, and the cited pages earned roughly 35% more organic clicks than the uncited ones.

One exception survived the decline. Amsive studied 700,000 keywords across ten sites. When a search already contained a brand name, the AI summary appeared for only a small share of those searches, and when it did appear, clicks on the results went up 18.68%. People who ask for you by name still click. Most buyers, though, don't start by typing your name. They start by describing their problem, and the answer to "who should I look at" is where the new competition lives.

What a devtool company can do about it

None of this requires a budget. It requires an afternoon and a spreadsheet.

First, find out whether AI names you. Write 15 to 30 questions a real buyer would type. Run them in fresh chats on ChatGPT, Perplexity, and Gemini. Log who gets named and which pages get cited. We published the full manual method, step by step (it takes one afternoon and no tools).

Second, read the pages that do get cited. They are usually not vendor homepages. They are comparison posts, Reddit threads, and docs pages that answer one question completely. If AI keeps citing a listicle that skips your product, that listicle is the page to get into. If it cites nothing current in your category, refresh the stale comparison pages you control and answer the category question directly on your own site, naming competitors the way buyers do.

Third, make your product easy to ask for by name. Brand searches were the one place clicks held up. Docs, changelogs, and comparison pages that state plainly what your tool is called and what it replaces give buyers the words to search for.

Fourth, re-run the test every month. One run is a photo. Three runs are a trend, and the trend is the thing you can manage.

A note on the numbers themselves

Studies in this area disagree on absolute numbers, badly. Measurements of the top result's click-through rate vary by a factor of twenty between research teams, because each team counts a different mix of searches. Everything quoted above is a relative change, measured by one team across time, because those hold up better. Dates matter too: every number here is from 2025 or 2026, and this market moves fast enough that any of them could look stale within a year. The sources are linked below; check them before you put any of this in a board deck.

Where we fit

We run a bigger version of this test for a living. CiteGap, the team behind this site, asks 20 to 30 buyer questions across the engines your buyers use, works out why your product is missing, and hands you a fix list, as a one-time project rather than a subscription. If you want a first look for free, the lite audit asks five questions about your product on one engine and emails you the summary.

Run the free lite audit

Sources