
Does SEO still work with AI answers?
On this page
- Myth 1: SEO is dead because search volume is collapsing
- Myth 2: AI Overviews reduce every site's clicks by 58%
- Myth 3: Being cited in the AI answer brings the traffic back
- Myth 4: Zero-click proves AI killed the click
- Myth 5: Ranking first guarantees an AI citation
- Myth 6: ChatGPT uses Bing so Bing rankings determine citations
- Myth 7: AI search requires llms.txt and special schema
- Myth 8: AI referrals have already replaced organic traffic
- Myth 9: Bot requests and citation dashboards measure AI visits
SEO
About 49%, by our arithmetic on Ahrefs' figures: that is how far position-one click-through rate fell on informational keywords without an AI Overview, from 0.076 to 0.039 between December 2023 and December 2025. Does SEO still work? Yes, for queries that still send clicks and for becoming a source AI systems retrieve. It cannot honestly promise informational clicks at the old rates.
Ahrefs measured an even steeper fall on keywords with AI Overviews, reaching 0.016. Its February 2026 study calculated a 58% shortfall against a forecast based on the non-AIO trend, not against an unchanged search results page.
These studies measure different events. A person clicking any traditional result, a position-one desktop CTR, and a brand's aggregate organic CTR are separate quantities.
| Study | Sample | Measurement window | Metric | Result | Main caveat |
|---|---|---|---|---|---|
| Pew Research Center | 900 US adults; 68,879 unique Google searches, including 12,593 with summaries | March 1 through March 31, 2025 | Visits with a traditional-result click | 8% with a summary; 15% without | Observational browsing panel; query types differ |
| Ahrefs, February 2026 update | 300,000 keywords, split equally between AIO and non-AIO informational keywords | December 2023 versus December 2025 | Position-one desktop CTR versus forecast | 58% below forecast on AIO keywords | Correlation; desktop only; forecast uses the non-AIO trend |
| Amsive | 700,000 keywords across 10 sites in five industries; CTR analysis on 10,000 ranking keywords triggering AIOs | Analysis published April 16, 2025; exact comparison dates are not stated in the study | CTR change | Average decline of 15.49% | Selected sites and keywords; brand mix changes the result |
| Seer, April 2026 update | 53 brands; 5,471,127 queries; 2.43 billion organic impressions | January 2025 through February 2026 | Organic CTR on AIO-present queries | 3.19% initially; 1.31% floor; 2.36% at the end | Observational account mix; some segments have substantial outlier effects |
Myth 1: SEO is dead because search volume is collapsing
The claim. AI assistants are replacing search engines, so investing in SEO means investing in a disappearing audience. Gartner's prediction of a 25% decline frequently appears as proof that this has already happened.
The evidence. Gartner's February 19, 2024 press release forecast a 25% reduction in search engine volume by 2026. A forecast is not a measurement, and the public announcement does not supply an independently observed outcome.
Google presented a different picture on July 22, 2026. In Alphabet's second-quarter earnings remarks, Sundar Pichai said AI Mode was adding incremental Search queries, Search usage reached an all-time high during the World Cup, and AI features were sending billions of website clicks weekly. Those are Google's own claims, without an independent volume figure that settles the Gartner comparison.
Search volume and website traffic can also move in different directions. More searches can coexist with fewer outbound clicks per search; your analytics dashboard cannot tell you which happened across the entire market.
What to do. Judge SEO against qualified demand and business outcomes on your site. Check whether commercial queries still produce inquiries, whether branded searches reach the correct page, and whether important service pages remain indexed.
My judgment for a company of 10 to 200 people: abandoning those assets because of a forecast is a poor trade. A shrinking click opportunity calls for stricter page selection and measurement, not an obituary.
Myth 2: AI Overviews reduce every site's clicks by 58%
The claim. Take last year's organic traffic, subtract 58%, and call the remainder your AI-adjusted forecast. The number looks precise enough to put in a budget spreadsheet.
The evidence. Ahrefs' February 4, 2026 study measured desktop CTR across 300,000 keywords. Position-one CTR on AIO keywords fell from 0.073 to 0.016 between December 2023 and December 2025; the non-AIO informational comparison fell from 0.076 to 0.039.
Ahrefs used the comparison group's trend to forecast an AIO-keyword CTR of 0.037. The observed 0.016 was 58% below that forecast, an association that does not isolate every cause of declining clicks.
Amsive's April 16, 2025 analysis reported a 15.49% average decline across its 10,000 analyzed AIO-triggering keywords. Non-branded keywords fell 19.98%, while branded keywords gained 18.68%; only 4.79% of its branded keywords triggered an AIO. These are selected-site results, not contradictory estimates of the same universal effect.
Seer's larger account dataset also shows movement over time. Across 53 brands and 2.43 billion organic impressions, AIO-present CTR reached a 1.31% floor in December 2025, then recovered to 2.36% in February 2026. No-AIO CTR was 3.16% and 3.82% in those respective months; the recovery did not erase the gap.
What to do. Separate queries by purpose before assigning a traffic expectation. The exposure table gives you useful starting hypotheses, with each percentage attached to its own population.
| Query group | Measured exposure | Practical implication |
|---|---|---|
| One- or two-word searches | Pew: 8% produced summaries in March 2025 | Check actual intent; short queries had less summary exposure |
| Searches of 10 or more words | Pew: 53% produced summaries | Long questions are exposed, even when the wording is valuable |
| Queries beginning who, what, when, or why | Pew: 60% produced summaries | Treat answer-seeking traffic cautiously |
| Branded queries | Amsive: 4.79% triggered AIOs; CTR rose 18.68% on those that did | Measure branded demand separately |
| Informational queries | Seer: 36% AIO prevalence during its study window | Model lower click yield and citation exposure |
| Commercial queries | Seer: 8% AIO prevalence | Inspect comparisons and service-selection queries individually |
| Transactional queries | Seer: 5% AIO prevalence | Protect conversion paths; low prevalence does not mean immunity |
Seer's uncited transactional segment fell from 4.17% CTR to 2.15% during 2025, a 48% decline, then returned to 4.21% in February 2026. That rebound is a useful warning against treating one bad month as permanent market structure.
Semrush's study of more than 10 million keywords found informational queries' share of AIO queries fell from 91.3% in January 2025 to 57.1% in October 2025. Other intent categories gained share, so keep checking commercial exposure as well.
Myth 3: Being cited in the AI answer brings the traffic back
The claim. Once your page appears as an AI Overview source, the lost organic clicks return through the citation. Citation visibility becomes a substitute traffic target.
The evidence. In Pew's March 2025 browsing study, links inside summaries received clicks on about 1% of all visits to results pages that showed an AI summary. Traditional results received clicks on 8% of summary-present visits, versus 15% without summaries.
Seer's 2025 full-year informational averages show a different, complementary measure: overall organic CTR was 2.07% when the brand was cited, 0.94% when an AIO appeared without citing it, and 3.35% without an AIO. Those figures do not measure clicks specifically on citation links.
The uncited average also has a large caveat. One account supplied 47% of that segment's impressions at 0.20% CTR; excluding it raised the uncited average to about 1.61%. Seer explicitly cautions: “We cannot claim causation. Higher-authority brands are also more likely to be cited.”

The source-link path is narrower than the organic-result path. Treat this as a conceptual funnel, not a proportional chart or a claim that every non-clicking searcher stops reading.
What to do. Model citation visibility as a separate outcome, then price the remaining click opportunity. Say a hypothetical page receives 40,000 monthly informational impressions, with impressions held constant in every scenario.
monthly_impressions = 40,000 no_AIO_CTR = 3.35% cited_AIO_CTR = 2.07% uncited_AIO_CTR = 0.94% uncited_AIO_CTR_excluding_outlier = 1.61%
no_AIO_clicks = 40,000 * 0.0335 = 1,340 cited_AIO_clicks = 40,000 * 0.0207 = 828 uncited_AIO_clicks = 40,000 * 0.0094 = 376 uncited_clicks_excluding_outlier = 40,000 * 0.0161 = 644
cited_minus_uncited = 828-376 = 452 no_AIO_minus_cited = 1,340-828 = 512
This model gives citation-associated upside of 452 clicks against the uncited average, while the cited scenario still trails no-AIO by 512 clicks. These are scenarios using observed segment averages, not a causal forecast for your page.
Build content that earns snippets and citations when it also serves a business purpose. A useful source page can explain a buying constraint or document a technical decision; its value need not depend entirely on a citation click.
Myth 4: Zero-click proves AI killed the click
The claim. Roughly two-thirds of searches now end without a click, therefore AI Overviews caused roughly two-thirds of your lost opportunity. That quietly turns a search outcome into an attribution model.
The evidence. SparkToro estimated that 68.01% of US Google searches ended without a click between January and April 2026. Its Similarweb panel covered desktop and mobile browsers, excluding the Google app.
Earlier estimates were about 49% in 2019 using Jumpshot and 60.45% in 2024 using Datos. SparkToro itself describes comparing these panels as “a bit of apples and oranges”; the figures establish that zero-click behavior predates widespread AIOs, but they cannot provide a clean causal time series.
Semrush's paired-keyword analysis supplies a useful correction. Among keywords without an AIO in May 2025 that had one in October 2025, zero-click decreased from 33.75% to 31.53%.
That result does not prove AIOs increase clicks everywhere. It shows why a broad zero-click percentage cannot substitute for a matched analysis of affected queries.
What to do. Record both impressions and clicks for stable query groups. A falling CTR alongside rising impressions describes a different business problem from falling impressions and stable CTR.
For each group, inspect the result page and ask what still requires a website: a quote, a purchase, a detailed specification, or evidence the answer cannot supply. Allocate effort according to that remaining task, with informational pages supporting it where useful.
Myth 5: Ranking first guarantees an AI citation
The claim. Win the original Google query and the AI answer must cite you. The opposite claim, that rankings have stopped mattering entirely, is equally convenient and equally incomplete.
The evidence. Ahrefs examined 863,000 keyword SERPs and four million AIO URLs for its March 2, 2026 study. It found 37.9% of citation URLs within the top 10 SERP-block positions, 31.2% in positions 11 through 100, and 31.0% beyond 100; those positions count all result blocks, including ads, featured snippets, and People Also Ask.
Its blue-links-only test returned 37.1% in the top 10, 26.2% in positions 11 through 100, and 36.7% beyond 100. Keeping the definitions visible prevents a misleading comparison between a conventional organic rank and a position across every SERP feature.
The earlier top-10 figure was about 76% in July 2025. Ahrefs suggests fan-out and Gemini 3 as possible explanations, and notes it also improved its parsing in between, so part of the change may be measurement (our reading).
Google documents that AI Overviews and AI Mode may issue multiple related searches through query fan-out. A page can therefore become relevant to a supporting search even when it does not rank prominently for the original wording.

Follow the branches from the original question to supporting searches. A cited page can enter through one of those searches without occupying the original query's top 10.
What to do. Make pages eligible for retrieval, then cover the supporting questions your buyers actually ask. Google's eligibility requirement is specific: a page must be indexed and eligible to appear in Search with a snippet.
Give a service page concrete information about implementation requirements, limitations, and selection criteria. Create a separate supporting page when the question needs its own substantial answer; repeating the same paragraph across twelve pages does not add evidence.
Myth 6: ChatGPT uses Bing so Bing rankings determine citations
The claim. ChatGPT search is effectively a Bing results wrapper. Get into Bing's leading results and assistant citations should follow predictably.
The evidence. Seer's February 6, 2025 study examined roughly 100 questions and more than 500 SearchGPT citations. More than 87% matched Bing's top 20 organic results for the same question; its authors called the sample limited and the findings directional, with questions skewed toward top/best recommendations.
Ahrefs tested 15,000 long-tail prompts using data from early July 2025. ChatGPT's in-text citation URLs overlapped Bing's top 10 by 8.1% and Google's top 10 by 8.0%; across the studied assistants, average overlap was about 10% for Bing and 12% for Google.
These studies differ in prompt selection, date, and ranking cutoff. They cannot establish a clean change in provider behavior, but the larger study contradicts treating same-prompt Bing rank as a dependable citation predictor.
OpenAI's ChatGPT search FAQ says search sometimes partners with other search providers and links privacy policies for Microsoft and Shopify. Providers include Microsoft; OpenAI has not disclosed the mix, so the strong claim that every answer comes from Bing exceeds the available evidence.
What to do. Verify Bing indexing because it is a sensible retrieval prerequisite. Then inspect actual citations for a representative set of buyer questions, recording the prompt, date, answer, and cited URL.
Treat mentions as evidence of visibility, with causation kept separate. Ahrefs' May 26, 2025 study of 75,000 brands with Domain Rating above 40 found branded web mentions had the strongest measured correlation with AIO brand mentions, a Spearman coefficient of 0.664, versus 0.218 for backlinks.
That is correlation, not a purchasing instruction for synthetic mentions. Earn references through material someone can check, such as documentation or a defensible comparison; Google's own guidance cautions against inauthentic mentions.
Myth 7: AI search requires llms.txt and special schema
The claim. Conventional SEO needs a new technical layer: an llms.txt file, special AI schema, and tiny content chunks. Without that layer, assistants supposedly cannot understand your pages.
The evidence. Google's AI optimization guide, updated July 10, 2026, says Google Search does not use llms.txt. It also says structured data is not required for generative AI search and there is no requirement to break content into tiny pieces.
Google's AI features documentation states there are no additional eligibility requirements or special optimizations for AI Overviews and AI Mode. That is Google's stated behavior for its own search systems, not a universal declaration about every assistant.
Structured data can still describe supported entities and qualify pages for applicable Search features. That useful role does not establish a separate schema requirement for AI citations.
What to do. Check crawl access, indexing, snippet eligibility, and whether the rendered page contains the answer you expect a system to retrieve. Keep structured data accurate and consistent with visible content.
Use headings that identify the question being answered and keep necessary context beside the answer. The full SEO and AI-search method connects those checks to page selection and coverage; a ceremonial text file cannot repair missing product information.
Myth 8: AI referrals have already replaced organic traffic
The claim. ChatGPT and other assistants now send enough traffic to justify moving most SEO resources into assistant visibility. A fast-growing channel is assumed to be a large channel.
The evidence. Ahrefs measured 74,752 sites in March 2026 and found AI chatbots collectively sent 3.5 million visitors, 0.28% of total traffic. ChatGPT supplied 2.7 million, about ten times any other AI platform in that dataset.
Conductor measured 1.08% AI referral share across 1,215 US enterprise domains from May through September 2025. ChatGPT accounted for 87.4% of those AI referrals; the enterprise sample and earlier window make it a different benchmark, not a replacement estimate.
These figures exclude a large classification complication. Google says AI Overview and AI Mode clicks appear within Search Console's Performance report under the Web search type, so chatbot-referral share does not capture all visits influenced by AI answers.
Google's AI features documentation also says clicks from AI Overview results pages are higher quality, meaning users are more likely to spend more time on the site. That definition does not establish your business's lead quality or conversion rate.
What to do. For a company of 10 to 200 people, I would order the work by existing demand and the cost of neglect:
- Verify the pages that sell. Confirm important service and product URLs are indexed, accessible, and lead to a working inquiry or purchase path.
- Split performance by intent. Separate branded, commercial, transactional, and informational queries before setting targets.
- Repair exposed information pages selectively. Keep pages that answer real buying questions; add missing constraints and supporting evidence.
- Cover supporting searches. Build substantial answers to relevant subquestions that retrieval systems may use.
- Measure citations independently. Track where the company appears without translating every appearance into a visit.
- Expand assistant work when outcomes justify it. Compare qualified inquiries and sales contribution alongside referral volume.
If informational pages have lost clicks and your commercial pages and subquestion coverage need rebuilding, Iron Goo's SEO service is aimed at that work. Define the needed pages and measurement criteria before agreeing to a publishing volume.
Myth 9: Bot requests and citation dashboards measure AI visits
The claim. A rising ChatGPT-User count means more people arrived from ChatGPT. A citation chart or AI impression report can then stand in for the missing referral data.
The evidence. Google announced Search Console's Search Generative AI performance reports on June 3, 2026 and rolled them out to all sites on August 31, 2026. Google lists impressions as the metric, broken down by pages, countries, devices, and dates, and says more metrics may be added over time. Our reading of that list is that the reports include no clicks or query data.
Microsoft announced Bing Webmaster Tools AI Performance as a public preview on February 10, 2026. Microsoft describes citations, cited pages, and grounding queries, with sampling applied to grounding-query data and no click metric.
Server logs introduce another distinction. In irongoo.com's own logs, from October 10 through early October 24, 2026, 206 requests contained ChatGPT-User; six matched OpenAI's published IP ranges and 200 did not.
Of those 200 unmatched requests, 192 returned 404 responses, mostly while probing for secret configuration files. The same window contained 186 Perplexity-User requests: none matched Perplexity's published list of four prefixes dated October 17, 2025, and 182 returned 404s while probing configuration paths.
Verify assistant request addresses against the provider's published IP ranges. Keep verified fetches separate from human referral visits; a matching address establishes the source of a request, not a person arriving on your site.
This is one genuine request from that window, with an address matching OpenAI's published range. It records an assistant fetching a page, without establishing that a human visited it.
9.129.49.6 - - [12/Oct/2026:22:19:01 +0000] "GET /guides/marketing/marketing-in-ai-search-and-assistants HTTP/1.1" 200 34814 "-" "Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; ChatGPT-User/1.0; +https://openai.com/bot"
The first field is the client IP; the timestamp is UTC, followed by the HTTP request. Status 200 records success, 34814 is the response byte count, and the referer field is "-"; the final field contains the ChatGPT-User/1.0 product token.
What to do. Start in Search Console's Performance > Search results report. Compare monthly CTR and impressions for question-shaped queries using ^(what|how|why|when|who|is|does|can)\b, then compare branded and commercial groups separately.
That expression is a filter, not a complete intent classifier. Use query segmentation and decay diagnosis to investigate a falling line before attributing it to AI, and use the generative AI report for page-level exposure.
For an IPv4 nginx access log, this script checks claimed ChatGPT user fetches against OpenAI's published ranges. Run it with the log filename as the first argument; IPv6 clients require extending it to handle ipv6Prefix.
import ipaddress, json, re, sys, urllib.request
nets = [ipaddress.ip_network(p["ipv4Prefix"]) for p in
json.load(urllib.request.urlopen("https://openai.com/chatgpt-user.json"))["prefixes"] if "ipv4Prefix" in p]
real = fake = 0
for line in open(sys.argv[1], errors="ignore"):
if "ChatGPT-User" not in line:
continue
ip = ipaddress.ip_address(line.split()[0])
if any(ip in n for n in nets):
real += 1
else:
fake += 1
print("genuine:", real, "spoofed:", fake)
Our run on that irongoo.com log window printed genuine: 6 spoofed: 200. These counts demonstrate spoofing and fetch classification; they establish no human referral total.
For human referrals, parse the Referer URL and match its host, such as chatgpt.com or perplexity.ai. Searching the entire log line for the platform name can count an internal URL containing utm_source=chatgpt.com; keep referral sessions, verified assistant fetches, and AI impressions in separate columns.
Put your next steps in order
Build on what you have just read with a free ebook on semantic SEO and AI visibility. Get worked examples, decision rules, checklists and a do-this-week action in every chapter, plus a six-week plan for your site.

