AI & Automation

97% of People Also Ask Answers Are Now AI-Generated: Google Just Shut the Cheapest Traffic Door for SMBs

2026.09.10 · 118 views
97% of People Also Ask Answers Are Now AI-Generated: Google Just Shut the Cheapest Traffic Door for SMBs

AlsoAsked's 19.2M-query sample shows 86% in August and 97% in the first week of September, up from 12.6% fourteen months ago. Q&A content did not die; its delivery address changed.

Share:

On 9 September 2026, Barry Schwartz of Search Engine Roundtable published a chart that quieted the SEO industry. Mark Williams-Cook of the PAA tool AlsoAsked ran a back-test across a 19.2 million query sample of 2026 English searches and found that answers inside Google's People Also Ask (PAA) block were 86% AI Overviews-generated in August, jumping to 97% in the first week of September. A second vendor, Allintitle, put it more bluntly: its Also Ask Miner data shows 100% since August. Fourteen months earlier, the same metric sat at 12.6%.

PAA is not new. It launched in 2015 and for a decade served as the cheapest traffic door available to small businesses: you did not need a first-page ranking, only a cleanly structured paragraph that answered the question directly, and the collapsible panel might quote you and hand over a click. That mechanic is why every "FAQ page SEO" tutorial written in Taiwan over the past five years looks identical. The turning point came once AI Overviews rolled out broadly. Google wired retrieval-augmented generation into its core ranking systems, and PAA shifted from "quoted source text plus one outbound link" to "a model-rewritten paragraph plus several system-selected links." Semrush's analysis of more than 10 million keywords found PAA co-occurs with AI Overviews at 90.03%. The two modules have long run on the same machinery; August 2026 is simply when the last 14% of verbatim excerpts got absorbed.

The tooling market around this is splitting rather than consolidating. Question-mining veterans like AlsoAsked and AnswerThePublic sell question trees at US$11–59 per month: cheap, but they only tell you what people ask. Full platforms like Semrush and Ahrefs start around US$129 per month and add AI visibility modules that tell you who gets cited. Newer entrants such as Allintitle productise "is this PAA answer AI-generated" as a metric in its own right. Question mining is commoditising into a low-price tier while AI citation monitoring becomes the new high-margin tier. Whoever owns large-scale crawling gets to reprice.

For Taiwanese SMBs the implication is concrete: the FAQ page you have maintained for three years will probably hold its impressions and lose its clicks. This article solves three problems — what to measure once impressions and clicks decouple, what shape content needs to take to survive inside a generated answer, and how to monitor all of it without paying for a SaaS subscription.

What Happened, With the Full Numbers

Two independent datasets point to the same conclusion, though they disagree at the margin, which is worth noting. Allintitle's Also Ask Miner claims Google PAA answers have been 100% AI-generated since August 2026. AlsoAsked's 19.2 million query back-test returns 97%, with Williams-Cook stating it seems highly likely we reach 100% shortly. The month-over-month delta is the real signal: 86% in August to 97% in the first week of September, an 11-point jump in roughly four weeks. That is not gradual diffusion; that is a batch switch.

Zoom out and it gets starker. In mid-2025, Search Engine Land cited AlsoAsked's 8.4 million result sample showing AI-generated answers at just 12.6%. Fourteen months, 12.6% to 97%, roughly a 7.7x increase.

The click-side damage has third-party benchmarks. Ahrefs analysed 300,000 keywords and measured a 34.5% drop in position-one CTR when an AI Overview was present. Amsive's 700,000-keyword study found an average decline of 15.49%, with non-branded terms at -19.98% — but branded terms gained 18.68%. Same direction, very different magnitudes, because industry mix and query intent differ enormously.

The good news is that measurement became free this year. Google shipped Generative AI performance reports in Search Console and completed the global rollout to all sites by the end of August, exposing impressions from AI Overviews, AI Mode and Discover. The limitation is equally clear: it gives you impressions, not a click breakdown by AI surface.

Immediate Actions for Three Audiences

Brand owners and SMB operators

  • ☐ Accept one thing first: the KPI for an FAQ page moves from organic clicks to branded query volume plus direct traffic. Those are the metrics that register the downstream effect of AI impressions.
  • ☐ Shift budget from "write ten more FAQ articles" to "three pieces with first-party data" — real price bands, tested results, case numbers. A model cannot paraphrase those away; it has to cite you.
  • ☐ Audit branded terms. Both the Ahrefs and Amsive datasets show branded queries benefiting in AI environments. It is the one door still widening.

Marketers and SEO practitioners

  • ☐ Open the Generative AI performance report in Search Console and overlay the impression curve since July 2026 against GA4 organic sessions. Produce a list of pages where impressions rise and clicks fall.
  • ☐ Rewrite that list first: put the answer in the first paragraph under the H2 and finish it in 40 to 60 words, then expand below. RAG retrieves self-contained passages, not whole articles.
  • ☐ Kill programmatic Q&A content. Google's own documentation states there are no additional requirements or special optimizations for appearing in AI features. Stacking question keywords no longer offers leverage.

Developers and agencies

  • ☐ Complete clients' FAQPage structured data and verify the JSON-LD is not HTML-escaped. Its effect on AI citations is contested (see below), but it still works for traditional rich results and costs almost nothing.
  • ☐ Use a Laravel scheduled command plus the Search Console API to land generative AI impressions in a database daily. More reliable than manual CSV exports.
  • ☐ Price "AI visibility baseline measurement" as a separate deliverable in proposals rather than burying it inside an SEO retainer.

Question Mining and AI Visibility Tools Compared

ToolPositionMonthly price (list, Sept 2026)Best for
AlsoAskedPAA question-tree miningUS$15 / 29 / 59 (annual 12 / 23 / 47), 3 free searches dailyFast question maps for a single topic; cheapest entry point
AnswerThePublicQuestion and preposition keyword expansionFree (3/day); paid from about US$11, Pro US$99Early-stage content ideation; broader, shallower data
SemrushFull platform plus AI visibility modulePro from about US$139.95Competitor citation comparison, multi-market
AhrefsFull platform plus Brand RadarLite from about US$129Backlinks and AI mentions in one view
Google Search ConsoleOfficial generative AI performance reportFreeYour own site's impressions; the most accurate first-party data

The takeaway is blunt: paid tools sell you other people's data, while Search Console gives you your own for free. SMBs should exhaust the free one first.

What Nobody Will Tell You

Counterpoint one: 97% does not mean your traffic drops 97%. Expanding a PAA panel was always a heavily zero-click behaviour; most users read and collapse it. What actually changed is link sourcing — from one quoted page to several system-selected pages. For sites that were never quoted in the first place, this is a lowered barrier, not a catastrophe. Equating "PAA went AI" with "traffic goes to zero" is panic marketing.

Counterpoint two: the data comes from companies that sell PAA tools. PAA mining is the core product for both AlsoAsked and Allintitle, so this finding doubles as their launch material. More importantly the sample is entirely English-language queries. AI Overviews coverage and trigger conditions differ in the Traditional Chinese market, so transplanting 97% onto a Taiwanese site will mislead you. Measure your own; do not import conclusions.

Counterpoint three: schema's effect on AI citations is overstated. Vendor research has reported that adding JSON-LD to pages already appearing in AI Overviews produced no measurable citation lift (that study's sample and methodology are not fully public, so treat it as a signal, not a verdict). Google's documentation says the same thing in different words: there is no AI-specific optimization. Do schema, but do not treat it as a GEO silver bullet.

The Zero-Subscription SMB Alternative

  • Week 1 — Baseline. Export the Search Console generative AI performance report to CSV and compare it against the standard Google Search report for the same period. Compute an impression-to-click ratio per page. This is your only free, first-party AI visibility baseline.
  • Week 2 — Automate. Use the Search Console API (free quota) to write a Laravel Command that pulls page x query x impressions x clicks into MySQL daily, with a scheduler and a 7-day moving average. Roughly 200 lines, written once, used indefinitely.
  • Week 3 — Visualise. Connect MySQL to Looker Studio (free) and build a red-flag list of pages where impressions climb while clicks fall.
  • Week 4 — Manual sampling. Pick 20 core questions, search them in an incognito window, and record whether the PAA answer is AI-generated and whether your domain is cited. Twenty questions take about 40 minutes and beat any tool's Traditional Chinese coverage.

Four-week cost: zero currency, roughly 12 hours of labour. That is a complete AI visibility monitoring system.

FAQ

Now that PAA answers are AI-generated, will my site still get linked?

Yes, but through a different mechanism. Previously your text was quoted with one link attached. Now the model rewrites the answer and Google's systems select several supporting links themselves. Google's documentation describes AI features as RAG-based, showing clickable links that support the response, with selection controlled entirely on the system side.

Should I remove the FAQ section from my site?

No. An FAQ block remains the content shape most easily retrieved by RAG, and FAQPage structured data still works for traditional search results. What needs to change is density: stop answering common knowledge and start supplying figures only you have, such as real price bands, lead times and failure cases.

Is Taiwan at 97% too?

There is no public Traditional Chinese sample. All 19.2 million queries were English, and Google's search experience already differs by region. A reasonable inference is that Taiwan trails by some months, but the only reliable method is sampling 20 to 30 of your own core questions manually.

Do I still need an AI visibility monitoring SaaS?

Usually not, if you are a single brand in a single market. Paying makes sense when you need competitor citation benchmarking, coverage across multiple language markets, or third-party reports for clients or investors. If you only want to know whether you are being seen, Search Console is enough.

My Take

The consensus reading is that AI-generated PAA is another traffic apocalypse. My judgement runs the other way: this is structurally good news for small businesses, and the AI visibility monitoring SaaS category will lose most of its market to free Search Console functionality within 12 to 18 months.

The reason is that the barrier moved direction. Old PAA was winner-take-all — one question, one excerpt, one site collecting the impression, and that slot usually belonged to an already-authoritative domain. The new RAG mechanism selects several supporting links at once, which opens a gap for small sites that lack domain authority but carry specific first-party information. What actually dies is content-farm FAQ that paraphrases common knowledge, which should never have existed.

On the SaaS side, the timeline speaks for itself: Google introduced generative AI performance reports in June and finished the global rollout by late August. First-party visibility data is becoming free infrastructure. The only defensible moat left for paid tools is competitor data and cross-engine comparison, not your own numbers. That moat is too narrow to support current valuations.

For ScriptWalker the productisation opening is specific: package "AI visibility baseline setup" as a one-off project — Search Console API integration, MySQL persistence, a Looker Studio dashboard and a 20-question manual sampling report, all wired through Laravel scheduling and handed over for the client to run themselves. Clients skip an annual SaaS renewal, we earn on build plus follow-on content rewriting, and that is a far healthier business than reselling subscriptions. To discuss it:

Sources

Share:
AI & Automation Back to Blog