SEO Tips: What Actually Works When AI Is Eating Your Clicks

By Tom Morgan · First published May 9, 2026 · Fully revised October 31, 2026 · 12-minute read

If you only have a minute: AI Overviews still cost most sites clicks, but the 2026 data is less bleak than the 2025 headlines, and Google says the fix is ordinary SEO. Pages cited inside an AI Overview earn far more clicks than pages that are not. So the job is to rank, get cited, and measure both. This guide covers six tactics that hold up against the evidence, the order to fix things in, and the places where the data still disagrees.

What changed since the May edition

• The click story moved. Seer Interactive’s 2026 update found that organic click-through rate (CTR) on AI Overview queries bottomed out in December 2025 and then rebounded. Ahrefs data still shows declines. Section 1 lays out both.
• Google published official guidance (May 15, 2026). Its position: optimizing for AI Overviews and AI Mode is still SEO, and no special schema, llms.txt or AI-only rewrites are needed. I removed the earlier claim that FAQPage schema works as a “citation package,” because Google says otherwise.
• Search Console now reports AI impressions. The Generative AI report launched June 3 and reached all sites on August 31. My earlier line that citation gaps are “invisible” is only half true now.
• Two core updates are done, plus four spam updates. The latest core update finished June 2, and the September spam update finished October 8. The old “March update is rolling out” language is gone.
• Corrections. “Cited restores 35% of lost traffic” was wrong. The Core Web Vitals section also got a correction: LCP, not INP, is the metric most sites fail.

The numbers at a glance

Finding	Figure	Source
Organic CTR on AI Overview queries, June 2024 → September 2025	1.76% → 0.61% (reported as −61%; the raw rates imply about −65%)	Seer Interactive v2, 3,119 queries, 42 organizations
Same metric, December 2025 → February 2026	About 1.3% → 2.4%	Seer Interactive 2026 update, 53 brands, 5.47M queries
CTR by result type (Seer’s 2026 data)	No AI Overview ≈ 3.3%, cited ≈ 2.1%, not cited ≈ 0.9%	Search Engine Land, April 24, 2026
Position-1 CTR when an AI Overview appears	−58%	Ahrefs, 300,000 keywords, Dec 2023 vs Dec 2025
Position-1 CTR, July 2026 data	−52% (20.02% → 9.69%)	Advanced Web Ranking, 1.75M+ US SERPs per month
US Google searches ending without a click	68% in early 2026, up from 60.5% in 2024	SparkToro/Similarweb, as summarised by Advanced Web Ranking

1. The click gap is real, but it isn’t a straight line

Most “best SEO tips” lists still assume that ranking first means capturing a third of the clicks. That assumption is out of date. Where an AI Overview appears, the top result gets far fewer clicks.

Three independent studies agree on direction and disagree on size:

• Seer Interactive tracked organic CTR on AI Overview queries falling from 1.76% to 0.61% between June 2024 and September 2025. Its 2026 update, covering 2.43 billion impressions, found a low of about 1.3% in December 2025 and a recovery to about 2.4% by February 2026. Seer called the rebound surprising, and it still has not closed the gap with queries that have no AI Overview.
• Ahrefs measured position-1 CTR 58% lower when an AI Overview shows, and its December 2025 data suggests the gap widened. Seer itself notes that Ahrefs’ result differs from its own.
• Advanced Web Ranking put the position-1 drop at 52% using July 2026 data, with 74.3% of AI Overview results ending in zero clicks.

Why do they differ? They use different samples (client Search Console data versus keyword panels), different query mixes and different time windows. None of them is wrong, but none is a universal law either. Budget as if clicks on informational queries stay below their 2023 level, and don’t assume the rebound continues.

The finding that matters most for strategy is the one all the studies share: being cited helps. In Seer’s 2026 data, cited pages earned more than double the CTR of uncited pages on the same results, and an earlier Seer version found 35% more organic clicks and 91% more paid clicks for cited brands.

One correction from the May edition. That 35% means cited brands get 35% more clicks than uncited brands. It does not mean citation wins back 35% of the traffic that was lost.

2. What Google itself says about “GEO”

Generative Engine Optimization (GEO) is the label for making content easy for AI systems to retrieve and cite. It sells well, so it helps to know what Google has actually said.

Google’s guide to generative AI features (May 15, 2026) and its “AI features and your website” documentation make five points:

1. AI Overviews and AI Mode are built on Google’s core ranking and quality systems, so SEO best practice still applies.
2. There are no additional requirements to appear in them.
3. You don’t need AI-specific text files, special markup, or Markdown variants, and Google Search doesn’t use them llms.txt.
4. There is no special schema.org markup for AI features. Structured data stays useful for rich results but isn’t required.
5. A page that isn’t indexed and eligible for a normal snippet can’t appear in AI features.

Google’s guidance covers Google Search only. ChatGPT, Perplexity, and Claude run their own retrieval, so use GEO as a measurement label if you like, but don’t treat it as a separate bag of tricks.

3. Six things that actually move results

Each tactic below carries an evidence label. Strong means a published Google rule or consistent findings across sources. Moderate means solid data with an unclear mechanism. Directional means plausible, with thin production evidence.

3.1 Measure citation, then close the gap

Evidence: Strong that citation matters; Moderate on how to earn it.

Since August 31, Search Console’s Generative AI report has been available to all sites. It shows how often your URLs appeared in AI Overviews, AI Mode, and Discover’s AI features, broken down by page, country, and date. It reports impressions only, with no clicks or CTR, and sites with too few AI impressions may not see it. An impression doesn’t confirm that you were linked as a source. For that, check the result yourself.

Here is a repeatable two-hour check I call the Cited-or-Not Audit:

1. Export your top 30 informational URLs by impressions from Search Console.
2. Open the Generative AI report and note which of those pages show AI impressions.
3. For each page’s main query, search it and see whether an AI Overview appears and whether your URL is listed among its sources.
4. Sort the pages into three buckets: cited, ranking but not cited, and not ranking in the top 10.
5. Work the middle bucket first. Google already trusts these pages enough to rank them. The usual fix is structural: put a direct, self-contained answer in the first sentence under the relevant heading, with a source and a figure where one exists.
6. Re-check after 4–8 weeks.

The bottom bucket is an ordinary ranking problem, so don’t expect AI tricks to solve it. Whether ranking is required for citation is itself unsettled: studies of AI Overview sources report overlap with the top 10 anywhere from roughly 38% to 76%.

Stop doing this: rewriting title tags “for AI.” Nothing in Google’s guidance suggests titles are how overviews pick sources.

3.2 Write answer-first, source-dense pages

Evidence: Directional.

Google’s systems use query fan-out, which means they break a question into sub-questions and pull from several pages. Pages that answer one sub-question clearly, early, and with a verifiable claim are easier to use. Early academic work on generative engines, from Princeton and IIT Delhi researchers, suggests that statistics and named sources raise retrieval rates. Production evidence is limited, so treat this as good writing that happens to be machine-friendly, and don’t chop pages into “AI-sized chunks.” Google specifically says that isn’t needed.

3.3 Build E-E-A-T as infrastructure

Evidence: Moderate.

Experience, expertise, authoritativeness, and trust are not a field you fill in. Three layers have to line up:

• On the page: a named author with real credentials, the reason the page exists, how it was produced, and sources for every number.
• In the markup: Article with a Person author, datePublished and dateModified, and an Organization with sameAs links.
• Off the page: the author’s work, bylines and mentions elsewhere. Schema pointing at an empty LinkedIn profile signals nothing, because the markup only points at evidence that has to exist already.

On most teams, these layers belong to three different people (content, developers, and PR), so assign one owner.

3.4 Pass Core Web Vitals, but know what they buy you

Evidence: Strong on thresholds; Moderate on ranking impact.

The thresholds haven’t changed since INP replaced FID in March 2024. Measured at the 75th percentile of real visits, LCP should be ≤ 2.5 seconds, INP ≤ 200 ms, and CLS ≤ 0.1. Google uses field data from the Chrome UX Report over a rolling 28 days, so Lighthouse scores are for debugging and not for grading.

My May edition said INP was the bottleneck. The data doesn’t back that. In the 2025 Web Almanac, 62% of sites passed LCP, 77% passed INP, and 81% passed CLS, and LCP is the main reason overall pass rates sit near 48% on mobile. INP is the hardest to fix once you fail it, because it measures every interaction in a visit and one slow handler can sink the score. Chat widgets and consent banners are common culprits.

Google’s documentation treats page experience as one input among many and says it still favors relevant content when the experience is mediocre. So pass the thresholds because they help users and conversion, and don’t expect them to rescue thin content.

3.5 Build topic depth, not keyword coverage

Evidence: Strong.

Across the 2023–2025 core updates, sites with clear topic ownership, logical clusters, and consistent internal linking held up better than sites ranking for hundreds of unrelated terms. Pages that try to serve every intent at once tend to lose to pages that serve one well. If you already rank in positions 2–5 for a target query, improve that page instead of publishing a second one that competes with it.

3.6 Keep your spam-policy hygiene clean

Evidence: Strong.

Google ran four spam updates in 2026, which makes it the busiest year for them since 2021. The most recent ran September 24 to October 8, taking 13 days and 16 hours. It was the longest of the four, and Google announced no new policy with it. Google says recovery from a spam demotion can take many months and that you can’t file a reconsideration request for an algorithmic one.

Audit against the written policies: scaled content abuse (including mass-produced AI pages published without real oversight), copied content, keyword stuffing, and doorway pages. Using AI to help is not the issue. What the policy targets is publishing at volume without adding value.

4. The 2026 update ledger

Dates	Update	Notes
Feb 5–27	Discover core update	Discover only, US English
Mar 24–25	Spam update	Shortest on record, under 20 hours
Mar 27–Apr 8	March core update	12 days; Google’s first broad update of 2026
May 21–Jun 2	May core update	About 12 days; practitioners called it heavier than March
June	Spam update	About two days
Aug 18–21	Spam update
Sep 24–Oct 8	Spam update	Longest of the year; some reported core-update-like effects

As of late October, trackers list the May update as the most recent core update. Google doesn’t publish a calendar, so check the Search Status Dashboard before you blame an update for a drop.

5. A sensible order of operations after a drop

Treat this as working practice, not a Google rule. I have not seen a controlled test of it.

1. Wait for the rollout to finish, then compare the full window. Week-one data is misleading because rankings move in waves.
2. Remove blockers first. Fix crawling, indexing, and field-data failures (Core Web Vitals). They are cheap to find, and clearing them makes later content changes easier to attribute.
3. Rebuild content quality on the pages that matter most: high-intent pages ranking 2–5, pages with AI impressions but no citations, and thin pillar pages.
4. Give it time. Google’s guidance says improvements can take months to register, so measure on a quarterly cycle.

My earlier version claimed that technical problems “add noise” that stops quality gains from registering. That’s an untested hypothesis. The practical reason for the order is simpler: if you change everything at once, you can’t tell what worked.

6. What this means for your role

Agency account managers. Report traffic next to AI-feature impressions from Search Console, cited share on your tracked queries, and conversions. Stop describing AI Overviews as a temporary disruption, since four studies now show a lasting gap even where one shows a rebound.

Marketing managers. Run the Cited-or-Not Audit this quarter instead of a full site audit, and revise existing pages instead of publishing duplicates.

Budget owners. Don’t cut content spend based on 2025 click data alone. Seer’s rebound and the cited-versus-uncited gap suggest that a cited page can hold value while a Search Console clicks chart looks flat. Set expectations that topic-cluster investment takes many months to show up, and add AI impressions and conversions to the dashboard before any budget decision.

7. What we still don’t know

• Whether the CTR rebound lasts. SEMrush sees recovery, and Ahrefs and Advanced Web Ranking see continued gaps. The studies can’t be reconciled yet.
• How many clicks AI impressions produce. Search Console reports impressions only.
• How stable citations are. Source lists in AI answers change, so a citation you have today isn’t guaranteed next quarter. Track it monthly.
• Whether other assistants behave like Google. ChatGPT and Perplexity use their own retrieval, and Google’s guidance doesn’t cover them.

Anyone offering a settled 2026 playbook is working from incomplete data. What holds up is a short list: rank, be citable, measure both, stay clean, and update the page when the evidence changes.

Frequently asked questions

Do AI Overviews reduce organic click-through rates?

Yes, in most studies. SEER measured a fall from 1.76% to 0.61% on AI Overview queries through September 2025, then a partial rebound to about 2.4% by February 2026. Ahrefs (−58%) and Advanced Web Ranking (−52%) found large position-1 declines. The size varies by method.

What is the difference between being cited and being mentioned?

A cited page is linked as a source. A mention paraphrases information without sending the reader to you. SEMrush data compares cited and uncited pages, and cited pages got more than double the clicks. Search Console impressions don’t separate the two, so check the result manually.

Is GEO different from SEO?

For Google Search, Google says no: AI Overviews and AI Mode run on its core ranking and quality systems, and there are no extra requirements. GEO remains a useful word for tracking visibility across other AI assistants.

Do I need llms.txt or special schema for AI Overviews?

No. Google says Search doesn’t use it, and no special schema.org markup exists for AI features. Keep structured data for rich results and entity clarity.

Is INP harder to pass than LCP?

It isn’t. In the 2025 Web Almanac, more sites passed INP (77%) than LCP (62%). INP can be harder to diagnose and fix once you fail it.

Sources

Figures were checked against these sources on October 31, 2026. Where I could only reach a secondary summary of a primary study, I say so.

1. Seer Interactive, “Google AI Overview Study – SEO & PPC CTR impact” (v2 through September 2025; v3 through February 2026).
2. Search Engine Land, “Google AI Overviews CTR shows early signs of recovery: Study”, April 24, 2026.
3. Seer Interactive, “Inc. cites Seer research”, October 21, 2026 (notes the disagreement with Ahrefs).
4. Ahrefs, AI Overviews and clicks (position-1 −58%; accessed via summaries of the February 2026 update).
5. Advanced Web Ranking, Organic CTR by position, 2026 (July 2026 data; also summarises SparkToro/Similarweb zero-click data).
6. Google Search Central, “AI features and your website” and the AI optimization guide (May 15, 2026), read via the documentation and press coverage.
7. Search Engine Journal, “Google Search Console AI Reports Rolled Out Worldwide”.
8. Search Engine Land, “Google March 2026 core update rollout is now complete”.
9. Search Engine Roundtable, “Google May 2026 Core Update Has Completed Rolling Out” and “September 2026 Spam Update Is Done”.
10. CoreWebVitals.io, “How to Pass Core Web Vitals” (2025 Web Almanac pass rates; secondary source).
11. Pew Research Center, “Google users are less likely to click on links when an AI summary appears,” July 2025 (8% versus 15%).