Provley is a tool that tracks what ChatGPT, Perplexity, Gemini and Claude say when customers ask them who to use, and we spend our weeks looking at those answers. Most falling traffic reports we see follow the same pattern, and the fix follows the same order: diagnose by page, repair the pages that matter, and track AI answers directly. The guide below is written for UK small business owners and marketing leads who lose enquiries, not advertising pageviews.
Why is our Google traffic dropping since AI Overviews appeared?
Google traffic falls after AI Overviews appear because the AI Overview answers the searcher's question at the top of the results page, so many people read the summary and leave without clicking any link. Your page can still rank, still be shown and still be correct, yet earn fewer visits because the question was settled before anyone reached the organic results.
Searches that end without a click were common before AI Overviews, because featured snippets, knowledge panels and map packs already answered simple questions directly. AI Overviews extend that to longer and more involved questions: how to do something, what a term means, which option suits a situation. Relevant Audience’s recovery guide of 7 January 2026 reports that zero-click searches reached 68% of all queries in early 2026 (source: Fix AI Overviews Traffic Loss: 2026 Recovery Guide).
Publishers felt the change first. According to The Digital Bloom's 2025 report, the median publisher lost 10% of traffic year on year in the first half of 2025. Small service businesses tend to notice the shift later and differently, as fewer enquiries arriving through blog posts and guides rather than an obvious collapse in visits.
The loss is uneven across a site. Pages that explain things lose the most, because explanations are exactly what an AI summary can compress. Pages where someone wants to act, such as booking, buying, getting a quote or finding your address, lose far less, because a summary cannot complete the task for them. Knowing which kind of page lost the clicks tells you whether you face a serious problem or an expected change in how people use search.
Google is also no longer the only place the question gets answered. Customers now ask ChatGPT, Perplexity, Gemini and Claude the same questions they once typed into Google, and those answers name some businesses and ignore others. Recovering traffic therefore means protecting the clicks that still come from Google and making sure AI assistants name your business when the click never happens.
How much have AI Overviews cut click-through rates, and which searches are hit hardest?
AI Overviews cut click-through rates sharply on the searches where they appear, and informational searches carry almost all of the loss. The Pew Research Center found, in its analysis of 22 July 2025, that clicks fall to 8% of visits when an AI Overview is present, against 15% without one; The Digital Bloom’s 2025 Organic Traffic Crisis report carries the same figures.
The searches affected are overwhelmingly questions. In its study titled 'AI Overviews Reduce Clicks by 34.5%', Ahrefs found, in the study it published on 17 April 2025, that 99.2% of keywords triggering AI Overviews are informational in intent. Relevant Audience's recovery guide adds that informational queries trigger overviews around 90% of the time, while transactional and navigational queries are rarely affected (source: Fix AI Overviews Traffic Loss: 2026 Recovery Guide).
Most of the large studies, including the Pew Research Center's analysis of real browsing behaviour, use US data and focus on publishers. Their direction is consistent: when a summary appears, people click less on the results below it, and the links inside the summary itself are rarely clicked unless the summary leaves something unanswered.
Exposure by type of search, from most to least affected:
- Hit hardest: definitions and explainers, where the summary gives the whole answer
- Hit hard: how to questions with a standard set of steps
- Hit hard: comparisons and 'which is better' questions, where the summary lists the options
- Hit moderately: cost questions, where the summary gives a rough guide but people still want a real quote
- Hit least: branded searches for your business name
- Hit least: transactional searches such as book, buy, hire or get a quote
- Hit least: navigational and local searches, such as opening hours or a provider near a named town
For a small service business, the way AI Overviews hit research searches hard and leave buying searches mostly alone is good news wrapped in bad news. The traffic you are losing is mostly early research traffic: people who wanted to understand a topic, many of whom were never going to contact you. The searches that turn into enquiries are mostly ones AI Overviews leave alone. The risk is that research is where buyers build their shortlist, and the AI summary now has a large say in who is on it.
The size of the drop on your own site depends on your mix of pages. A site whose traffic came mainly from a popular explainer will see a steep fall; a site whose visitors arrive on service and location pages may barely notice. Published averages are therefore a guide to direction, not a forecast for any single business.
Why are clicks falling when our rankings haven't changed?
Clicks fall while rankings hold because an AI Overview sits above the organic results and answers the query, so your position stays the same but fewer people scroll down to your link. In Search Console the telltale pattern is steady impressions and steady average position, with clicks and click-through rate falling over the same period.
Average position in Search Console measures where your link appears in the list of results, not where it appears on the screen. A top organic result can sit below an AI Overview, a block of sponsored results, a map pack and a set of related questions. On a phone, the searcher may never see the link at all, even though Google records an impression and a high position.
Picture a heating engineer with a well ranked guide explaining why a boiler loses pressure. The guide still sits at the top of the organic results. Above it, an AI Overview now lists the common causes and the steps to repressurise. The homeowner reads the list, then either fixes the boiler or decides to call someone, and in neither case visits the guide. Search Console shows the same position and similar impressions, with clicks well down.
Impressions can even rise during a traffic drop. Google's documentation says links shown inside AI Overviews are counted in Search Console alongside other results, so a page used as a source but rarely clicked looks, in the data, like a page that has suddenly become less appealing.
Other causes can produce flat rankings with falling clicks, and each is worth ruling out:
- A richer result now sits near yours, such as reviews, prices or a video, and draws the click
- Searchers' needs changed, so the same query now leads people to a different kind of page
- Your page title or description changed and no longer matches the query well
- Analytics tracking broke, often after a cookie banner or consent change, so visits happen but are not recorded; compare Search Console clicks with analytics sessions, since Search Console does not depend on your tracking code
The pattern that points to AI Overviews is specific. The drop is concentrated in informational queries, the affected pages explain rather than sell, average position is unchanged, and the timing matches AI Overviews appearing for those queries. When you search the affected queries yourself and see an AI Overview at the top, you have your answer.
How do we check in Search Console whether AI Overviews are really the cause?
Compare a recent period with the same period a year earlier in Search Console's Performance report, then break the change down by page, query, device and country to see whether the lost clicks sit on informational queries with stable positions. Google's own guidance on traffic drops makes the same point: work out where and when the drop happened before changing anything.
- Open the Performance report for search results and turn on clicks, impressions, average CTR and average position
- Compare a recent period with the same period last year, so seasonal demand cancels out
- Set the country filter to United Kingdom if most of your customers are here
- Sort the Pages tab by change in clicks and list the pages that lost most
- For each of those pages, open the Queries tab and note whether the queries are questions, definitions or how to searches
- Check the Device tab, because AI Overviews push organic links furthest down on mobile
- Search the biggest lost queries yourself in a private window and note whether an AI Overview appears and which sites it cites
Each cause of a traffic drop leaves a different fingerprint in the data:
| Cause | Impressions | Position | Clicks and CTR | Other signs |
|---|---|---|---|---|
| AI Overviews | Steady | Steady | Down | Mostly on informational queries |
| Ranking drop | Down | Down | Down | Often across the site close to a Google core update |
| Seasonality | Down for the same months last year | Not the signal | Down for the same months last year | Google Trends shows lower demand for the topic |
| Indexing problem | Fall away entirely | Not the signal | Fall away | Affected pages move from indexed to excluded in the Pages report |
| Tracking problem | Not the signal | Not the signal | Steady in Search Console | Analytics sessions down |
| Site change | Drop from one day | Drop from one day | Drop from one day | The drop starts on the day of a redesign, migration or change of web addresses |
| Manual action or security issue | Usually down across the whole site | Usually down across the whole site | Usually down across the whole site | Flagged in its own Search Console report |
Search Console does not, at the time of writing, offer a filter that separates searches showing an AI Overview from searches that do not. The diagnosis is therefore an inference from the pattern, backed by checking queries by hand. Keep a simple sheet listing each lost query, the date you checked, whether an overview appeared and which sources it named. The sheet becomes your starting list for fixes.
Avoid acting on a site-wide average. A site can lose clicks on a handful of old explainer posts while its service pages gain, and the average hides both movements. Every decision that follows should be made page by page.
How do we get our pages cited in AI Overviews, ChatGPT and Perplexity?
To get cited, make sure each AI system can reach and index your pages, then write pages whose answers can be lifted out and quoted on their own, backed by named sources and real depth on the topic. Being cited is now a separate goal from ranking: Slate reports that, according to Ahrefs, only 38% of AI Overview citations now come from pages on the first page of organic results (source: Google AI Overviews Statistics: CTR & Traffic Impact | Slate).
Technical eligibility comes first. Google's AI Overviews draw on pages that are indexed and eligible to show a snippet, so a page blocked in robots.txt, marked noindex or carrying a nosnippet rule cannot be quoted. ChatGPT, Perplexity and Claude use their own crawlers and search sources, and some hosting, firewall and bot protection settings now block those crawlers by default. Check your robots.txt file and your security settings for these crawlers:
- Googlebot: needed for Google Search and AI Overviews
- Bingbot: feeds Bing's index, which several AI search tools draw on
- OAI-SearchBot: lets your pages appear in ChatGPT search answers
- PerplexityBot: lets Perplexity find and cite your pages
- Claude-SearchBot: lets Claude cite your pages when it searches the web
- Google-Extended: controls use of your content for Gemini model training, not whether you appear in Google Search
Extractable answers come next. AI systems quote passages, not whole pages. A passage is easy to quote when the opening sentence under a heading answers the heading's question directly, the passage names its subject instead of leaning on 'it' or 'this', and any figure carries its source in the same sentence. Headings phrased as the questions customers actually ask help the system match a passage to a query. Lists and tables help wherever the content is a set of steps, options or comparisons.
Topical depth matters because AI systems prefer a source that covers a subject properly. A single thin post rarely gets cited. A cluster of pages that answers the main question and its follow-ups, linked to each other, gives the system several passages to choose from and signals that your business knows the field. For a service business, depth usually means writing down what your experts already explain to customers on the phone: the costs, the timelines, the exceptions and the mistakes people make.
Entity clarity is the part most small businesses miss. AI assistants recommend businesses they can identify with confidence. Use the same business name, category and location on your website, your Google Business Profile, your Companies House record, directories and review sites. State plainly on key pages what you do and where you do it. Organisation and LocalBusiness structured data help Google connect those facts, though markup alone does not earn a citation.
Mentions on other sites count too. ChatGPT and Perplexity often cite trade bodies, review sites, local press and comparison articles when they answer 'who should I use' questions. Being listed accurately in those places, and earning a mention in the articles assistants already cite for your topic, is often faster than waiting for your own page to be chosen.
Freshness helps on topics where rules or prices change. Show a visible updated date, revise figures when their sources update, and remove advice that no longer applies. Assistants that search the web live tend to favour recent, clearly dated pages for those questions.
Which of our pages are most exposed, and what should we change on each?
Explanatory pages are the most exposed: guides, blog posts, glossaries and how to articles that answer an informational question an AI Overview can summarise. Service, pricing, contact and local pages are the least exposed, so a small service business should turn its exposed explainers into the source AI cites and make its low exposure pages the ones that convert.
- Explanatory articles and guides, high exposure: put a direct answer in the opening sentences, add the detail a summary cannot hold, such as worked examples, UK rules and your own process, and link to the matching service page
- How to pages, high exposure: keep the steps in a numbered list, add the point where people usually need professional help, and say who you are and why you know
- Comparison pages, high exposure: use a table with clear criteria and say plainly who each option suits, including when your own service is not the right fit
- Service pages, low exposure but central to AI recommendations: state what you do, for whom, where, how long it takes and how pricing works, in plain sentences an assistant can repeat
- Pricing pages, low exposure: give real prices or clear ranges, say whether VAT is included and explain what changes the price
- Location and contact pages, low exposure: keep name, address, phone number and service area identical to your Google Business Profile
- Case studies and reviews, low exposure: name the problem, the type of customer and the outcome so assistants can match you to similar requests
Picking the page to fix first is a matter of value, not volume. A workflow that suits most small businesses:
- Take the list of pages that lost clicks from your Search Console check
- Mark each page with the service it leads to and whether it ever produced enquiries
- Drop pages that attract visitors with no link to anything you sell
- For the rest, run each page's main question through ChatGPT, Perplexity, Gemini and Claude and note who they name
- Choose the page where competitors are named and you are not, on a question tied to a profitable service
- Make the edit, record the date, and recheck the same question every week
The exact edit depends on why the page lets you down. When an assistant answers the question but cites a competitor, the usual reason is that the competitor states the answer in a quotable sentence while your page buries it beneath an introduction. Move the answer to the top, phrase the heading as the question, and name your business and location in the passage.
A vague or wrong answer from an assistant usually means your page lacks a specific fact the assistant needed, such as what a service costs, how long it takes or which UK rule applies. Add that fact plainly, with its source if it is not your own. When no assistant names anyone at all, the question is open, and a well structured new page is often enough to become the default answer.
Leave alone what still works. Pricing, contact and booking pages that convert should not be rewritten into long explanations to chase citations. Their job is to give an assistant or a searcher the facts needed to act: price, area, availability and how to get in touch.
New pages should also start from what AI assistants already say. Before writing, check what ChatGPT, Perplexity, Gemini and Claude say today for the question you want to own. If every answer repeats the same general advice, your page needs the specific detail only a working business knows: what usually goes wrong, what the job involves in practice, and what customers in your area tend to ask next.
How do we measure success when clicks no longer tell the whole story?
Measure success by how often AI assistants name your business when customers ask the questions you want to win, tracked every week, alongside Google clicks and enquiries. Google clicks show only the searches where someone left the results page; they say nothing about the answers where your business was recommended, or ignored, without any click at all.
Being named also feeds back into Google clicks. Slate cites Seer Interactive’s September 2025 update, which found that brands cited inside an AI Overview got about 35% higher organic CTR across the third quarter of 2025 (source: Google AI Overviews Statistics: CTR & Traffic Impact | Slate). A citation can therefore win back some of the clicks the overview took away, so tracking citations is not a vanity exercise.
The method for measuring how often AI assistants name your business is simple enough to run by hand:
- Write down the questions your customers actually ask before they buy, in their own words, including your town or region where it matters
- Run each question through ChatGPT, Perplexity, Gemini and Claude every week, in a fresh chat with no history
- Record whether each answer names your business, where in the answer it appears, which competitors are named and which pages are cited
- Work out the share of answers that name you across all the questions and assistants
- Note each question where you are missing and a competitor is named, and trace which page the assistant would need from you
- Keep the wording identical each week so changes reflect the assistants, not your phrasing
AI answers vary from run to run, so a single check proves little. The trend over several weeks is what matters: a rising share of answers naming you shows the edits are working even while Google clicks stay flat. Keep the question set fixed for a full quarter before changing it, otherwise the results cannot be compared.
Beyond the weekly check of AI answers, other signals show whether AI assistants are sending you customers. Analytics can show referral visits from ChatGPT, Perplexity, Gemini and Claude; the volumes are small but tend to convert well, because the visitor has already been recommended to you. Branded searches in Search Console often rise when assistants mention you, as people look you up by name. A 'how did you hear about us' question on your enquiry form, with an AI assistant option, captures the recommendations that never produce a trackable visit.
Report a short set of measures each month: your Citation Share (the share of AI answers naming you), Google clicks to service and contact pages, and enquiries by source. Together they show whether the business is being found, even when the old headline measure of total organic traffic keeps sliding.
Provley’s weekly AI check runs this for you, and the Monday email reports the result. Provley puts your customers' questions to ChatGPT, Perplexity, Gemini and Claude every week, reports the share of AI answers that name your business, and gives the exact edit for the page that lets you down. The questions and page checks stay fixed, so each week's result is comparable with the last. You can also connect Provley to ChatGPT or Claude and ask how your website is doing from the assistant you already use, and a Provley audit is available without opening an account.
Does any of this apply differently to UK businesses?
The mechanism is the same in the UK, but the evidence is thinner, because nearly all the headline studies use US data, so UK businesses should check their own Search Console figures rather than assume US rates apply. The best regional anchor available is Slate's report that AI Overviews appear on roughly 47% of informational queries in Europe (source: Google AI Overviews Statistics: CTR & Traffic Impact | Slate).
Several UK specifics change the practical advice. Filter Search Console to United Kingdom so overseas traffic does not blur the pattern. Run your AI checks with UK wording and place names, because assistants often default to US sources, spellings and rules when a question does not say where the asker lives. A question about tax, tenancy, employment or planning answered with US rules is a gap a UK page can fill.
UK rules and terms are an advantage when competing for citations. Pages that explain how something works under HMRC, Companies House, the FCA or your local council, written in UK spelling, with prices in pounds and VAT made clear, give assistants a UK source to cite where many competing pages are American. Local detail helps in the same way: the towns you cover, regional differences in cost, and the UK bodies your work answers to.
Entity details need a UK slant too. Keep your business name and address consistent across Google Business Profile, Companies House, Yell, Checkatrade, Trustpilot and the trade body directory for your sector, whichever apply. Assistants that recommend local providers often lean on exactly these listings.
Expect further change. Google keeps extending AI features across UK search, including AI Mode, so a page that holds its clicks in 2026 may lose them once an overview arrives on its queries. A weekly AI check and a regular Search Console comparison give early warning before enquiries fall.
Frequently asked questions
Should we block Google from using our pages in AI Overviews?
No. Blocking is rarely worth it for a small service business. A page marked noindex or carrying a nosnippet rule cannot be quoted in AI Overviews, and Slate cites Seer Interactive’s September 2025 update, which found that brands cited inside an AI Overview got about 35% higher organic CTR across the third quarter of 2025 (source: Google AI Overviews Statistics: CTR & Traffic Impact | Slate). Google-Extended controls use of your content for Gemini model training, not whether you appear in Google Search.