AirPulse
    ← Back to Insights
    AI TechnologyBrainstorm

    Info Edge just put AI search on the earnings call.

    Harsh Songra··

    Shiksha lost 13% of its billings to what management called AI-driven changes in search behaviour. Chegg lost 49% of non-subscriber traffic in a single month. The hack response is a trap. I will write what we learned running AirPulse on our own company, not what would sound good on a sales call.

    We saw the pattern on our own dashboards before it showed up in anyone’s earnings call. The same prompt to ChatGPT, Claude, Perplexity, Gemini and Copilot returning very different recommendation sets. Brands we expected to dominate the category quietly absent. Brands we had never heard of confidently named. The instinct was to dismiss it as noise. We measured it instead.

    This quarter, Info Edge made the shift legible on a P&L. Shiksha billings of Rs 45 crore for Q4 FY26, down 13% year-on-year, with management attributing the decline to AI-driven changes in search behaviour that have been pressuring Shiksha’s traffic for several quarters. Q4 was the quarter that pressure crossed from traffic into billings. The headwind, they told investors, will persist.

    I have been waiting for an Indian listed company to put it in this language. Not a marketing problem. Not a traffic problem. A billings problem. The plan Info Edge announced is the plan you make when you have decided this is not a blip. Domestic counselling capability so you stop relying on Google to deliver buyers. AI voicebots to scale that counselling. Expansion into the UAE and continental Europe. None of those moves get easier. They did them anyway.

    Figure 1. Reported declines attributed to AI search, across recent public disclosures. Periods vary; figures are year-on-year change at the time disclosed.

    I am writing this from a specific angle, so I should say it upfront. I am on the founding team at AirPulse. AirPulse measures what Info Edge just disclosed. How often AI engines mention and recommend a brand, where in the answer, with what sentiment. Our heaviest user is Airmeet, the company I work at every day. So when I read the Shiksha line, I read it next to the dashboards I look at every week. The argument in this post is not theoretical.

    Why we think Info Edge is the start, not the exception

    Indian listed companies do not put speculative explanations on earnings calls. The compliance discount on every word is high. When management names a cause on a public transcript, they are saying both that they believe it and that they expect questions about it for the next several quarters. The Shiksha disclosure passes both tests.

    The US version of this story is older. Chegg’s Q1 2025 revenue was $121 million, down 30% year-on-year, subscribers down 31%. The lawsuit they filed against Google in February 2025 disclosed that non-subscriber traffic had collapsed to negative 49% in January against negative 8% nine months earlier. Their CEO said publicly that AI Overviews was a significant factor. The market took roughly 90% of the cap off the company since IPO. The shape of the disclosure matters here. Public companies do not produce that level of attribution detail unless they have already accepted the underlying explanation internally.

    Stack Overflow is messier because Prosus does not break them out, but their own data is queryable and the trend is brutal. Monthly questions fell from 108,563 in November 2022, the month ChatGPT launched, to about 25,566 in December 2024. A 76% drop in user-generated activity on what used to be the dominant developer Q&A site. That is what the front edge of the curve looks like when a category resets.

    Then there are the publishers nobody hears about unless they were one of them. HouseFresh, the independent air-purifier review site, lost about 95% of its Google traffic in the September 2023 Helpful Content Update. Daily visitors went from roughly 4,000 to 200. They recovered in October 2025, but it took two years of work that had nothing to do with SEO. YouTube. Real editorial depth. That recovery is the thesis of this entire post.

    What the macro data tells us we should expect

    The cleanest CTR study I have read is the one Pew Research published in July 2025. 900 US adults, 68,879 Google searches, March 2025. When an AI summary appeared, click-through to a regular search result fell to 8%. Without a summary it was 15%. The click-through on links inside the summary was 1%. Session abandonment was 26% with the summary versus 16% without. Most of those buyers are ending their journey without a single click.

    Figure 2. Click-through behaviour with and without an AI summary in Google results, Pew Research, March 2025.

    Similarweb’s 2025 publisher report adds the macro number. News-site traffic from Google fell 26% year-on-year in the twelve months after AI Overviews launched. Zero-click searches climbed from 56% in May 2024 to 69% in May 2025. Thirteen points in one year, on one product surface.

    The number I cannot stop thinking about is from Cloudflare. They run a meaningful chunk of the public internet’s edge, so they can count what crawlers fetch and what referrals come back. In July 2025, Anthropic was crawling 38,065 pages for every one referred visit it sent to a publisher. OpenAI was 1,091 to 1. Google, for comparison, was 5.4 to 1. About 80% of AI crawling is for training, not real-time retrieval. The AI engines are taking value from your content at orders of magnitude more than they return. If your plan is to produce more pages and hope to catch more citations, you have just signed up for a trade that is thirty-eight thousand to one against you.

    Figure 3. Pages an AI crawler fetches for every one visit it sends back. Cloudflare Radar, July 2025.

    What we did at Airmeet

    Airmeet sells a virtual and hybrid events platform. Our category has been under pressure since 2023 for separate post-pandemic reasons, and that is not the story I am telling here. The story is that our buyers, marketers and event organisers, ask AI engines questions every day that we want to be the answer to. Best webinar platform for B2B. Airmeet vs Zoom for virtual conferences. How to host a hybrid event with 1,000 attendees. If we are not in the recommendation set across the five engines, the pipeline dries up regardless of what our marketing dashboards say is happening on Google.

    So in April 2026 we ran AirPulse on ourselves. The numbers are honest because they are ours.

    85 prompts that match how our buyers actually phrase research. 222 analysis jobs against the five engines. 23,558 citations tracked across them.

    Three things I learned, in the order I learned them.

    First, the same query gives different answers across engines. I had assumed this in theory. Watching it happen is different. ChatGPT, Claude, Perplexity and Gemini each have a different prior on what counts as the right answer to “best virtual event platform for large conferences.” Optimising for one is not optimising for the others. The work is engine-by-engine, not a single ranking exercise.

    Second, our starting visibility was low and uncomfortable to look at. On the virtual-events prompts we tested, our visibility scored between 5% and 20% depending on the engine and how we phrased the query. We were not being cited reliably. The instinct in marketing is to issue a release. The actual question is harder. Why does the model have no reason to mention us?

    Third, we thought about hacking it. Every team under pressure thinks about it. Generate more pages. Target more keyword permutations. Get the model to repeat your name through brute force. We did not, because Google’s spam policy is unambiguous on the question. Using generative AI tools to produce pages at scale without adding value is named as scaled content abuse. The fastest way to lose AI search visibility tomorrow is to try to game it today. HouseFresh is the named, datable proof of what hacking costs.

    What does work

    The framework AI engines have inherited is the one Google’s Search Quality Rater Guidelines describe. Experience, Expertise, Authoritativeness, Trust. Of those four, the September 2025 update calls Trust the most important. Untrustworthy pages rate poorly no matter how much experience, expertise or authority they appear to have. That is the rubric AI engines are reading against, even when they are not citing the rubric by name.

    There is no separate AI ranking system. When Claude does a web search, it queries a real search backend and emits citations against the results that backend returned. When ChatGPT cites a page, Bing’s index trusted it enough to surface. AI engines inherit the brand authority signal the search engines already use. There is no AI SEO channel. There is brand authority, and AI inherits it.

    At AirPulse, the four signals we look at first when we audit any brand, including Airmeet, are these.

    • Canonical entity naming. Pick one form of your brand name. Declare it in Organization JSON-LD with the legitimate variants in alternateName, and use the canonical form consistently in titles, bylines and owned media. Inconsistent naming fragments your mention count silently. We have agents in airpulse that help you generate and fix them.
    • Third-party validation surface. AI engines cite Reddit, Quora, analyst notes and vendor-comparison pages more readily than they cite owned blogs. The work is to be present and accurate on the surfaces the model already reads. Not to publish another blog on your own domain.
    • First-hand evidence in owned content. Founder POV, real customer numbers, screenshots, internal benchmarks, named examples. AI models can tell when a piece has no original contribution behind it. So can readers, increasingly. The bar is not to publish more. It is to publish what only you could have written.
    • Expert review on what ships. Every long-form piece from our team carries a named author and a named reviewer. If a claim has no source, it does not ship. This is the rule the AirPulse Content Quality Gate enforces in code. It is also the rule our editors enforce in practice.

    Figure 4. The AirPulse audit framework: four signals applied to every brand we work with, Airmeet included.

    These are unfashionable answers. They are slow. They do not produce a press release at quarter-end saying AI visibility is up 40%. They are also the only answers that have ever worked. AI engines did not invent a new trust currency. They are paying out in the one that always mattered.

    The counter-story

    Google would disagree with this whole post. In August 2025 Liz Reid, the Vice President of Search, published a blog arguing that organic click volume from Google Search is relatively stable year-on-year and that AI Overviews drive higher quality clicks. The post showed no underlying data. Pew, Similarweb, Cloudflare, Chegg’s lawyers, and now Info Edge measure the ground differently. Both stories can be true at the macro level. Total click volume stable while distribution shifts hard against publishers. But for any one publisher staring at their analytics, the shift is what they feel.

    It is also worth noticing which public companies are playing offence. Yelp’s Q2 2025 shareholder letter disclosed AI search API calls up 20x year-on-year and AI search data licensing already running at more than $10 million in ARR. Pearson, in its 2025 preliminary results, reported 6% operating profit growth and called AI a tailwind for its content products. The brands monetising the shift are the brands that did not wait for it to stop.

    What to do this quarter

    Assume Info Edge has read the room correctly. Assume the headwind persists. The brands that will look least bad on their next four earnings calls are the ones doing this work now.

    • Measure AI visibility separately from SEO traffic. They are different channels. Do not fold them into one chart. The KPIs that actually matter in the pre-click era are different.
    • Audit your Organization, Article and FAQPage JSON-LD on every owned page. It is the cheapest week-one investment.
    • Find the third-party pages that AI engines are already citing for your category. Reddit threads. Comparison pages. Analyst posts. Be present and accurate on them. Do not republish them on your own site.
    • Stop generating scaled content. If a piece would not have been written without the keyword permutation, do not publish it.
    • Put a real expert byline and a real expert reviewer on every long-form piece. AI engines reward this. Readers reward this. The Quality Rater Guidelines reward it explicitly.

    One last thing

    Info Edge will spend the next few quarters proving whether their pivot works. Chegg will spend them in court. The brands whose names will not show up in next year’s version of this post are the ones treating AI visibility as an authority problem, not a hacking problem. We are betting AirPulse is one of them. I would rather be wrong about that than right about the alternative.

    · · ·

    How this was made

    How this was made: the draft of this article was generated from AirPulse, our own AI engine optimisation platform, then reviewed and edited by the named author. Every external figure traces to a primary source linked below. Internal observations from running AirPulse on Airmeet are labelled as such.

    Sources

    FAQ

    Frequently Asked Questions

    01What did Info Edge disclose about AI search on its Q4 FY26 earnings call?
    Info Edge reported that its education portal Shiksha posted Q4 FY26 billings of Rs 45 crore, down 13% year-on-year, and management attributed the decline to AI-driven changes in search behaviour. It is one of the first Indian listed companies to name AI search on a public transcript as a billings issue rather than a marketing or traffic one, and it told investors the headwind will persist.
    02Why is AI search visibility a revenue problem and not just a marketing problem?
    When buyers research with ChatGPT, Perplexity, Gemini, Claude or Copilot and a brand is not in the recommendation set, the pipeline dries up regardless of how the brand ranks on Google. Public companies like Info Edge, Chegg and Yelp now report this on the P&L, which moves AI visibility from a soft marketing metric to a measurable driver of billings and revenue.
    03Does optimizing for one AI engine like ChatGPT improve visibility on the others?
    No. The same query returns different recommendation sets across ChatGPT, Claude, Perplexity and Gemini because each engine has its own prior on what counts as the right answer. Optimising for one does not optimise for the rest, so AI search visibility work has to be done engine-by-engine rather than as a single ranking exercise.
    04Does generating more content at scale help a brand get cited by AI engines?
    No, and it is actively risky. Google's spam policy names using generative AI to produce pages at scale without adding value as scaled content abuse, and sites that tried it, like HouseFresh, lost roughly 95% of their Google traffic. The fastest way to lose AI search visibility tomorrow is to try to game it with mass-produced content today.
    05How do you actually improve a brand's visibility in AI search results?
    AI engines inherit the same brand authority and E-E-A-T signals (Experience, Expertise, Authoritativeness, Trust) that search engines already use, with Trust weighted most heavily after the September 2025 Quality Rater Guidelines update. The four signals that move the needle are consistent canonical entity naming in Organization JSON-LD, presence on third-party surfaces AI cites (Reddit, Quora, comparison and analyst pages), first-hand original evidence in owned content, and a named author plus expert reviewer on every long-form piece.
    06How much are AI Overviews and zero-click search reducing traffic to websites?
    Pew Research found that when an AI summary appeared in Google results, click-through to a regular result fell to 8% versus 15% without one, and only 1% of users clicked a link inside the summary. Similarweb reported news-site traffic from Google down 26% year-on-year after AI Overviews launched, with zero-click searches rising from 56% in May 2024 to 69% in May 2025.
    Keep reading

    Related articles

    May 21, 2026

    Google I/O 2026: The Web Just Got a New Primary User, and It Isn’t Human

    What Google announced, what it actually means, and how your website has to change before agents start doing the clicking.

    Read article →

    January 8, 2026

    Definitive Source vs. Multi-Source Synthesis: What AI Search Really Wants in 2026

    The Short Answer: Being the single "definitive source" still matters for factual queries, but the practical win in AI search is becoming a non-negotiable building block of synthesized answers. AI engines like ChatGPT, Perplexity, and Google's AI Overviews prioritize consensus over dominance, meaning brands that position themselves as "necessary parts" of multi-source answers often achieve [...]

    Read article →

    September 9, 2025

    How Your LinkedIn Company Page Engagement Directly Impacts AI Search Visibility

    The B2B buying landscape has fundamentally shifted. With 70% of prospects now conducting initial research through AI chat assistants like ChatGPT, Perplexity, and Claude before engaging sales teams, your LinkedIn company page has evolved from a nice-to-have into a critical component of your AI search strategy. Here's the reality: AI systems are actively evaluating your [...]

    Read article →
    ← All Insights