Discussion starter: an AI search question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.
- AI Search
Has anyone watched AI citations rise while referral clicks stayed flat for months? At what point do you stop calling citations a leading indicator?
Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.
1 reply
-
adminAdmin
A rise in citations alongside flat referral clicks is what the available click data would predict, so I would not treat it as an anomaly. In Pew Research Center’s tracking of 900 US adults’ browsing in March 2025, users clicked a link inside Google’s AI summary in 1% of visits to results pages that showed one. A citation in an AI answer is closer to a footnote than a shop window: it lets readers check the source, and on Google’s AI summaries very few of them do.
There is also a timing problem in calling citations a leading indicator of referral clicks. The click happens when the answer is shown, not months later. If citations rise and clicks per citation stay low, you are looking at the click-through rate of the surface, not a delay. The only lagged route to more referral clicks is clicks per citation rising over time, which you can measure directly. Beyond that, citations can only lead outcomes that build up over time, such as branded search demand, direct visits or pipeline. I would stop calling them a leading indicator once three things are true: the rise is in your share of citations, not just in the number of AI answers; you wrote down in advance which outcome they should lead, by how long and by how much; and two of those lag windows have passed with no movement against a control. From then on, report citations as an exposure metric and judge them on whether that exposure is worth paying for.
Why rising citations with flat clicks is the expected pattern
Pew also found that users who saw an AI summary clicked a traditional search result in 8% of visits, against 15% when no summary appeared, and were more likely to end their browsing session (26% of pages with a summary, compared with 16% without). These are figures for Google’s AI summaries from US adults in early 2025, so treat them as direction rather than a benchmark for your market. The direction fits how these surfaces work: the answer is the product, and the cited page is supporting material.
The reporting makes the gap look wider, because the tools that count citations don’t count clicks:
- Search Console’s “Search generative AI performance” reports show impressions for AI Overviews, AI Mode and generative AI features in Discover, but no clicks, CTR or queries.
- Bing Webmaster Tools’ AI Performance report counts citations and shows sampled grounding queries, without clicks or traffic.
- GA4’s “AI Assistant” channel group collects referrals from assistants such as ChatGPT, Gemini and Copilot, though Google hasn’t published the full source list, and visits that arrive without a referrer still land in Direct.
- Clicks from AI Mode arrive in GA4 as google / organic, according to practitioner reports, so they sit inside organic search rather than in any AI channel.
So “referral clicks stayed flat” may really mean “the clicks we can attribute stayed flat”. Check that before drawing conclusions in either direction.
Confirm the citation rise is real
- Share, not count. Raw citation counts rise when platforms show AI answers on more queries, even if your share of those answers is unchanged. In June 2026 Bing announced a Citation Share metric, rolling out in preview, calculated as your site’s percentage of all citations shown for the same grounding query. That is a better series than totals. For prompt-tracking tools, use visibility percentage across repeated runs rather than a single run.
- Intent mix. The same Bing update classifies grounding queries into intents such as informational, commercial and navigational. A rise concentrated in definitions and how-to questions has little click potential by design. A rise on commercial queries that produces no visits is the more meaningful warning.
- Reporting changes. New reports, rollouts and tool method changes create steps in the data. Search Console’s AI reports reached all eligible sites worldwide on 31 August 2026, and a tracker that changes its prompt set can shift your “visibility” overnight. Mark those dates before reading any trend.
Decide what citations are supposed to lead
A leading indicator needs three things written down before you look at the results: the outcome it predicts, the lag you expect, and the size of effect that would count. Without them the claim can’t fail, and a metric that can’t fail is comfort, not evidence. The lags below are starting assumptions to replace with your own.
Outcome Where to measure it Lag to test Known gaps AI referral sessions GA4 AI Assistant channel, by landing page None: clicks happen alongside the citation No-referrer visits land in Direct; AI Mode clicks sit in google / organic Clicks per citation over time Referral sessions divided by citations, per page and platform Months, if familiarity builds Misleading if you divide one platform’s clicks by another platform’s citations Branded search demand Search Console branded queries filter Weeks to months PR, advertising and seasonality move it too Direct visits to deep, cited pages GA4 landing pages within the Direct channel Same period Noisy proxy: bookmarks, email and apps land there too Self-reported source “How did you hear about us?” with an AI assistant option on forms, surveys and sales calls Your sales cycle Recall bias and small samples Pipeline or revenue CRM, joined to the sources above Your sales cycle Multi-channel journeys blur attribution The one form of the leading argument that could apply to clicks is familiarity: people who keep seeing your brand cited might become more likely to click it later. If that is happening, clicks per citation on the same platform should rise over time. Months of flat clicks per citation are evidence against it.
A rule for demoting citations
Is the rise in citation share, not just raw count? If not, fix the metric first.↓Were the outcome, lag and minimum effect written down before you looked? If not, write them now and restart the clock.↓Have two lag windows passed? If not, keep watching, and call citations a hypothesis rather than an indicator.↓Did any of the outcomes you wrote down move more than a control did?↓Yes: keep citations as a leading indicator for that outcome only, with the lag you observed. No: reclassify them as an exposure metric.Why two windows: one window can be noise, seasonality or a coincident campaign. Two consecutive windows without movement, against a control that behaved the same way, is a reasonable point to stop claiming a link. It doesn’t prove no link exists. It shows the link is too weak to steer budget.
The control matters as much as the window. Useful controls are comparable pages or topics that aren’t cited, markets where the brand isn’t visible in AI answers, or the period before the citation rise. Watch for selection bias: pages that get cited may be stronger pages anyway, so compare their trends before the rise as well as after it.
Flat can still be a good result
Compare cited pages with similar uncited pages over the same period. If the uncited pages lost clicks while the cited pages held steady, citations may be protecting traffic in a shrinking click pool. That doesn’t make them a leading indicator, but it changes the business case from “more traffic later” to “less loss now”, which is a claim you can defend with data. Check the pre-period trends first, for the selection reason above.
Hypothetical example:
A B2B software company tracks 150 grounding queries in Bing. Its citation share rises from 6% to 14% over six months, and most of the gain sits in informational intents. GA4 AI Assistant sessions stay at about 400 a month, branded impressions are flat year on year, and about three demo requests a month name an AI assistant in the “how did you hear about us” field, unchanged. The team had predicted branded impressions would rise within a quarter. Two quarters later nothing has moved against the control markets, so citations move to the exposure section of the report. One follow-up stays open: the 20 commercial grounding queries, where share rose most, get a separate test on the pricing and comparison pages they cite.
What to report once citations are demoted
- Report AI visibility as exposure: citation share by intent and platform, with run counts or sampling notes.
- Report AI referral sessions separately, with their known gaps, and never as a function of citations.
- Put the budget question to the business directly: is being the cited source worth paying for as brand or reputation exposure, even without clicks? That can be a legitimate yes for some organisations and a clear no for others.
- Keep the commercial-intent subset under test, because that is where a link to clicks is most plausible.
If I were reporting this, I would drop the claim that referral clicks will catch up with citations unless clicks per citation are actually rising, because each click arrives with its citation. I would keep the “leading” label only for outcomes that can plausibly lag, with the lag written down, and give it two windows against a control. Past that point, citations measure exposure. Exposure can be worth paying for, but it should be budgeted as exposure, not as traffic that hasn’t arrived yet.
Need help with this on your own site?
Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.
Add a reply Cancel reply
Replies are open to approved members. Log in or apply to join.