The State of Organic and AI Search for B2B SaaS - Q3 2026 Update

Google hasn't lost its traffic to AI. Most searching still happens on Google, and that has not changed.

What changed is where your buyer starts. Before they visit anyone's website, many of them now ask a chatbot which vendors they should look at. Whatever it answers becomes their shortlist.

So the question is no longer "how do we get more visitors." It is "when someone asks ChatGPT who to consider in our category, are we on the list?" If you are not, you lose deals you never hear about, because the buyer never got as far as your website.

82% of people who buy business software have asked an AI chatbot for a vendor recommendation in the last two years.

And 69% of them ended up choosing a different vendor than the one they had in mind, because of what the chatbot said.

Those come from two G2 surveys of about a thousand software buyers each — one in March 2026 and one in June. They are buyers describing what they did, not a software company guessing.

Run the ten-minute test this week, before you commission anything. Ask ChatGPT, Google’s AI Mode and Perplexity what they recommend in your category. That one answer tells you more about your position than any dashboard you could buy.

Nothing Google did in August should change your plans.

The thing worth your attention has nothing to do with any single change: ranking well and getting visitors are no longer the same thing. Companies are holding their positions in Google and losing traffic anyway, because Google now answers the question on its own page.

If your marketing reports still open with "we rank third for X," they are reporting a number that stopped predicting revenue.

Google made one confirmed change in August: a routine anti-spam clean-up that ran from the 18th to the 21st. Google told reporters there was nothing new in it.

Its last significant change to how pages rank ended on 2 June. That is the longest quiet stretch this year, so another is probably close — and Google will not give notice.

The tools that watch Google for sudden movement flagged unexplained turbulence on seven separate days in July and August. None of it was ever explained. What site owners keep reporting, in their words: "When keyword rankings improved, but traffic decreased by 14%."

Google's free reporting tool will now tell you when your pages turn up inside an AI answer, but it won't tell you whether anyone clicked, or what they asked.

So you can see that you appeared, but you can't see whether it was worth anything.

The report arrived on 3 June 2026. It counts appearances only. No clicks, no questions. It also mixes Google's two different AI formats together, so you can't tell them apart.

There is a bigger hole underneath. When someone clicks from Google's chat-style AI through to your website, Google removes the information that would tell your analytics where they came from. Those visitors show up as if they typed your web address from memory (direct).

Google has begun routing clicks through a middle address (google.com/goto) that outside software can't read. It is aimed at companies scraping Google, including AI firms.

The side effect lands on you: the rank-tracking software your agency reports from may stop working. Nothing to do today, but if your reporting gets patchy this autumn, this is why — and it is one more reason to report on results you own rather than positions you rent.

Reported on 26 August. A rank-tracking firm called Nozzle measured it at "nearly a 100% rollout" after about four months of quiet testing.

Two honest caveats. The journalist called it confirmed but published no statement from Google. And nobody yet knows how badly it breaks the commercial tools — including some that produced data used in this report.

If an agency offers you a recovery plan for "the late-August update," ask which update. There is not a confirmed one.

No respected analyst has published a winners-and-losers study, and no source anywhere named business software as affected.

The industry's main news site published nothing about turbulence on 22, 23, 24 or 25 August. Four quiet days is the strongest evidence available that the period was calm. The story came back on 26 August, but only as one journalist saying that instability "suggests" something unannounced.

The firms that normally publish this analysis — Sistrix, Amsive, Glenn Gabe, Marie Haynes — published nothing at all. Sistrix did cover the March change, so the silence looks deliberate.

The best theory going, and it is guesswork rather than data, is that the clean-up hit websites that were automated from day one, and left alone established sites that later started using AI to help. One operator put it well: "It's not about whether it was created with generative AI, but what it was created for."

Don’t chase the turbulence — there is nothing to chase. Spend the energy on changing what your marketing report opens with. If it still leads with positions in Google, ask for branded search volume and pipeline that touched your content instead.

The web is not losing all its clicks. A smaller group of winners is taking them, and the dividing line is simple:

• Named inside Google's AI answer? You get roughly double the clicks of the companies that are not.
• Not named? You lose about half your clicks on your own category terms — to a competitor who is.

So the costly outcome is not Google showing an AI answer. It is Google showing one that leaves you out.

Two findings, both solid.

Researchers at the Indian School of Business and Carnegie Mellon ran the only proper experiment in this field — real people, real browsers, randomly split into two groups. When Google shows an AI answer, clicks through to websites fall 39.8%. That applies to the searches where an AI answer appears, not to your whole website.

Separately, the agency Seer Interactive looked at 5.5 million searches. Being named inside the AI answer produced about 120% more clicks than being left out. And on pages where Google showed no AI answer at all, clicks have actually gone up around 30% since early 2025. That is what "concentrating" looks like.

Write down the twenty questions that decide deals in your category. Check, by hand, whether you are named in the AI answer for each one. That list is your priority order, not your ranking report, and not a vendor’s visibility score.

Almost every statistic published about AI search is measured across the whole internet — mostly consumer questions your buyers never ask. B2B SaaS searches are hit considerably harder.

The searches losing the most traffic are the two your team has probably spent the most on: "best [category] software" and "[you] versus [competitor]."

Which points to one budget decision. Do not cut content spending, change what you buy with it. Visitor numbers are the wrong goal now. Being present in the answer, on the twenty or thirty questions that decide your category, is the right one.

The clearest measurement of business software specifically comes from Overthink Group, which has tracked a thousand software search results every month since March. As of June, more than 76% of searches where somebody is comparing solutions now return an AI answer.

Head-to-head searches are worse. 95.4% of "X versus Y" searches trigger one, the highest of any kind.

Google is also pushing further into exactly the stage where business buying happens. Semrush looked at 600,000 search terms and found AI answers on comparison-type searches grew 71% in six months, while falling 5% on searches where someone is about to buy.

Paid advertising is affected too. In Seer's study of six industries, business software came off worst: ads were seen 51% more often while the share of people clicking them fell from 12.8% to 10.3%. More exposure, fewer visitors, same budget.

"Reddit is taking over search results." True in 2024. Not now — forum posts have shrunk to a small fraction of business software results.

"Nobody ranks any more, so SEO is dead." Also wrong. Ranking well in ordinary Google is still the most reliable way into the AI answer.

On Reddit: forum posts fell from 9.5% to 3.9% of what appears on a business software results page between May and June. Google said in August that Reddit gets no special treatment. Reddit's own chief executive called its search traffic "choppy and volatile" on the last earnings call, and the share price fell on it.

On SEO: Seer found a strong relationship between ranking on Google's first page and being mentioned in AI answers — the strongest of anything they tested, with links from other sites barely registering. Google's own guidance, updated 10 July, says it plainly: "optimizing for generative AI search is optimizing for the search experience, and thus still SEO."

Protect the content budget and redirect it. Fewer posts, more depth, aimed squarely at the comparison-stage questions. If your team is measured on how much they publish, change the measure before you change the plan.

For years the standard play in business software was to publish lots of pages titled "best [category] software" and put your own product first. Google has now started pushing those pages down.

Volume stopped working, evidence started working. One question separates the companies that gained from the ones that lost a third of their visibility: how many of these pages do you have, and could a stranger tell you actually tested anything?

• Risky: dozens or hundreds of near-identical pages, your product ranked first in all of them, no real testing behind any.
• Still works: a small number of genuinely researched comparisons — including ones where a competitor beats you.

Ask your marketing lead for a count. If it is over about twenty and nobody can point to the research behind them, your next quarter is a clean-up, not a publishing push.

Four findings. They look contradictory until you line them up.

• In February, Lily Ray found software companies running between 61 and 340 of these self-ranking pages had lost 29–49% of their Google visibility after a January change.
• ChatGPT did start quoting list-style pages less often — but only a little, and during a stretch when it cut all its quoting by roughly a quarter.
• Google's AI answers moved the opposite way on the same questions. So "AI stopped quoting these pages" was never true across the board. As recently as June, roundups still appeared on every "best X" question tested.
• The biggest study yet, published by Kevin Indig in late August, found 62% of vendor-written list pages actually gained traffic, with a typical gain of 38%.

Put together: Google did not punish the format. It punished a particular abuse of it — publishing them at scale with nothing behind them.

Count your “best software” pages this month. If it is over about twenty and nobody can point to the research behind them, that is a liability sitting on your domain. Consolidate them into a handful you would be happy to defend publicly — including where a competitor wins.

Ask your marketing lead one question: "Which AI are we measuring?"

If the answer is a single overall score, that number is hiding something. We checked five competitors in one software category, and who looks like the winner completely reverses depending on which AI you ask.

A single blended score can tell you that you are winning while you are losing the one your buyers actually use.

On 27 August 2026 we used Ahrefs' database to count how often six different AIs quote each company — ChatGPT, both of Google's AI formats, Gemini, Perplexity and Copilot, in the US.

"Quoted" means the AI linked to a page on that company's website while answering somebody's question. It measures how often you get referenced. It is not traffic and it is not revenue.

The category is work-management software — Asana, monday.com, Smartsheet, Wrike and ClickUp. Five companies chasing the same buyer.

Build this exact table for your own category. It takes an afternoon with any AI visibility tool that reports per engine — and if yours will only give you one blended score, that is your answer about the tool.

Then ask two questions of it. Which engine do our buyers actually use, and are we winning it? A competitor level with you on the total may be beating you 2 to 1 where it counts. And what share of our visibility sits in a single AI? Anything above roughly half is a concentration risk, because that AI can change its behaviour overnight.

The split that matches the evidence: fund your own website for questions about you, and fund third party websites for questions about the category.

Most software companies fund only the first, which is why they never come up when a buyer asks an open question about the market.

This is where it is easy to draw the wrong conclusion, so here it is plainly.

Across the whole internet, Reddit and YouTube get quoted thousands of times more than any software company (our own Ahrefs pull, US, 27 August 2026). That tempts people into "our website doesn't matter." It is the wrong read — Reddit's millions of mentions are overwhelmingly about things that have nothing to do with software.

Narrow it to actual software buying questions and it reverses. Kevin Indig looked at 35,000 ChatGPT mentions on software questions only and found company websites hold 66.7–71.8% of mentions at every stage of the buying process — with community sites at 17.1% and news publishers at just 4.0%.

Both are true. Which one you are looking at depends entirely on what you counted.

Two things. Before you buy any AI visibility tool, demand the engine-by-engine breakdown — if a vendor will only show you one blended score, it is hiding a reversal like the one above.

And on where to show up: be where your buyers are, whether or not an AI quotes it. If your buyers live in a particular subreddit, community or LinkedIn feed, be there because they are there. Being quoted by AI is a bonus on top, not the reason to go.

Short answer: no — and do not create a new job title for it either.

There is no evidence that a separate "AI search" team beats a good search, content and PR team pointed at the right targets.

If someone is pitching you this, one question does most of the work: "What percentage of this is standard SEO, content and PR?" Honest answers land around 70–80%. Anyone claiming it is an entirely new discipline is selling you a new name for work you may already be paying for.

First, the industry can't agree what to call it. Searches for "generative engine optimization" peaked in July 2025 and have fallen 66% since (our own Ahrefs pull, US, 27 August 2026). A rival term, "answer engine optimization," overtook it. For scale, "seo agency" still gets ten times more searches than either.

Second, the job never got created. Semrush looked at 3,900 American job adverts and found these terms appear in just 6.3% of senior postings. Two years in, the work has been absorbed into existing search, content and PR roles.

One agency founder put it to Digiday this way: "If a GEO service does not openly tell you that success in AI visibility is 80% good fundamental SEO, they are selling snake oil."

Do not open a new headcount line or sign a new agency just for this. Ask the search, content and PR people you already pay to add “are we named in AI answers for our category?” to what they already report on. If they can't, that is a capability conversation with your existing supplier, not a reason to buy a second one.

The boring things work. The exciting things don't.

Rank well. Get into other people's comparison articles. Keep your pages current. Get talked about on websites you do not own. That is most of it, and your existing marketing team or agency probably already knows how to do all four.

Everything sold as a shortcut has either failed a proper test or was never properly tested.

The most careful work in this field is an academic review published in July 2026 that went through 45 separate studies. Its conclusion, in the authors' words: "No reviewed technique shows a stable, longitudinal, cross-platform causal effect."

In plain terms: once an AI has found your page, you can make it a bit more likely to get quoted. Almost nothing reliably gets you found in the first place, which is why ranking in ordinary Google still matters more than any AI-specific trick.

Two entries in the table below changed this quarter. Ranking dropped from "enough on its own" to "necessary but not enough." And publishing your own "best software" pages went from tactic to liability. A playbook written before February gets both wrong.

Pick the three boring ones and fund those first: rank properly for your category terms, update your top pages every quarter, and get into other people’s comparison articles. They are unglamorous, they are the best evidenced, and your existing team can start on Monday.

Keep AI visibility out of your board pack as a forecast.

Every AI in this report moved by double digits — sometimes by 90% — inside a fortnight, without warning or explanation, in the last nine months. Google at least announces its big changes. The AI companies announce nothing.

Track it as an early warning. Never attach a revenue number to it, and treat any single month's swing as noise until a second source confirms it.

A partial list from the last nine months, all measured by credible firms:

• Gemini went from naming its sources in 99% of answers to 76% — in two weeks. Individual companies lost between 12 and 92 percentage points of visibility.
• ChatGPT cut its quoting by 86–94% across five countries between February and April, then reversed most of it in May.
• Google's AI answers went from drawing 76% of their sources from the top ten search results to 37.9% in a year.

And this is not a measuring problem. Profound tested it directly — 753 questions across seven platforms, comparing 129,000 measurements against 860,000. Measuring ten times a day instead of once cut the noise by about 40%, and what was left was real movement by the platforms, not measurement error.

Treat AI visibility as weather, not climate. Watch it monthly, never forecast on it, and wait for a second source before reacting to any single swing.

The Reddit lesson generalises: if your buyers are in a community, be in it because they are there. Whether an AI credits that community this month is a decision made by two companies you have no relationship with, and it can reverse in a week.

Two decisions you can personally make this quarter:

• Put "how did you hear about us?" on your demo form. An afternoon of work, and currently the only thing that catches the buyer who read about you in ChatGPT and typed your name into Google three days later.
• Before renewing any AI visibility tool, make them answer the industry's new disclosure questions in writing — which AIs they check, how many questions they ask, how often they refresh. It is a published standard rather than your opinion, and a lot of vendors will fail it.

Your analytics understates AI's influence by roughly ten times, while your branded-search line quietly takes the credit.

Similarweb followed AI recommendations through to real visits: only 8.8% arrive looking like they came from an AI. Another 55.9% arrive as somebody searching your company name, days later.

Profound put a harder number on it using two million AI conversations matched to real browsing. After a company is mentioned in an AI answer, people visit that company's website at 1.5 to 2.5 times the normal rate over the next seven days — and more than 97% of those visits carry no tracking information at all.

The standard to hold vendors to: on 3 August the advertising industry body published "Measuring Visibility in the AI Era". It splits the problem into four questions — does your company appear, where and how prominently, in what context and accurately, and does it actually drive action — and separates rough directional numbers, which it says are not good enough for budget decisions, from properly disclosed ones.

Add the “how did you hear about us?” field to your demo form this quarter and start collecting. You cannot fix attribution backwards — whatever you do not start measuring now, you will not be able to answer questions about in six months.

One item needs a decision from your company, and it has a deadline: check who is allowed to read your website, before 15 September.

You are not a newspaper. Your content exists to win customers, not to be sold. For almost every software company the right answer is to let the AI crawlers in — being in the training data is how you become the default recommendation in your category. Blocking it to protect content you give away free applies a newspaper's economics to a software company's problem.

Five-minute conversation with whoever runs your infrastructure. Just do not let a default setting decide it for you.

On 15 September, Cloudflare — which sits in front of a large share of the web — changes its default for newly added websites to block AI training and agent crawlers on pages carrying adverts, while still allowing search crawlers.

Two wrinkles. A crawler that does more than one job gets blocked for all of them, which can catch Google's own crawler. And Cloudflare's new Bot Preference Sync, launched 21 August, now writes these choices into your website's crawler instructions automatically.

Everything else below is worth knowing but needs nothing from you.

Book fifteen minutes with whoever runs your infrastructure before 15 September and make an explicit decision about AI crawlers. For almost every software company that decision is “let them in.” It is the only dated action on this entire list.

Ranked by strength of evidence and speed to impact, not by what is easiest to sell.

Move 01 · This week

Ask the ten-minute question yourself

Open ChatGPT, Google AI Mode and Perplexity. Ask each "best [your category] software for [your ICP]." Note whether you appear and who does. If you're absent from all three, that's your Q4 priority and you found it without buying anything.

Move 02 · This week

Change what your marketing report leads with

Rankings no longer predict traffic, and traffic no longer predicts pipeline the way it did. Lead with branded search volume, AI citation presence on your top category questions, self-reported attribution, and pipeline from organic-influenced accounts.

Move 03 · This week

Add "how did you hear about us?" to every form

With 8.8% of AI-influenced visits trackable and AI Mode stripping referrer data entirely, self-report is the only instrument that sees zero-click influence. An afternoon of work. Don't wait for the industry to solve attribution — it hasn't.

Move 04 · Before Sept 15

Audit your crawler policy deliberately

Decide, explicitly, whether AI crawlers can read your site. For most B2B SaaS the answer is yes to all three categories — search, agent and training. The risk isn't choosing wrong; it's a CDN default choosing for you.

Move 05 · Q4 · Do the audit first

Clean up your "best X" pages, then rebuild comparisons

Count how many "best [category] software" articles on your site rank your own product #1. Companies with dozens of these lost 29–49% of organic visibility after Google's January change. Consolidate or rewrite them with real evidence. Then invest in a small number of honest head-to-head comparisons — still the highest-citation format there is.

Move 06 · Q4

Split your content budget by question type

Take next quarter's content plan and label every item either about us or about the category. If the second column is empty, that's your gap — and it can't be filled on your own blog. Category answers get won on roundups, analyst and media comparisons, YouTube and LinkedIn.

Move 07 · Before you renew

Put your AI-visibility vendor through the IAB filter

Demand decision-grade disclosure in writing: engines covered, prompt library construction, sample size, refresh cadence, reproducibility. If nobody clears the bar, that's a legitimate answer — and the money goes to Move 06.

How this was put together

Compiled 27 August 2026 by Kalungi. The citation and search-demand figures we call “our own” were pulled from the Ahrefs database on 27 August 2026, for the United States. Everything else was traced back to the original study and dated. Where two studies disagree, we show both rather than averaging them. We excluded statistics from sites that aggregate other people’s numbers without sourcing them, and vendor case studies with no before-and-after. Where a number could not be traced to its source, we cut it — including several that appear in competing reports.

Every claim above links to where it came from. If we have got something wrong, email me at cris.cubero@kalungi.com and I will correct it.