Chapter 4: Not All Platforms Are the Same

Last updated: November 1, 2026.

“We show up in ChatGPT” tells you very little about how you look in Google’s AI Mode or in Copilot. Every platform follows the chain from Chapter 3: split the question into searches, retrieve pages, read the relevant passages, write in its own words, cite sources. But which index they read from, when they decide to search, where they put the link and what data they give site owners differ quite a bit.

For Google AI Overviews and AI Mode, ChatGPT, Gemini, Copilot, Perplexity and Claude, it’s enough to look at the same three things: where the sources come from, how the link is shown, and what a site owner can see and control.

This is the part of the book that will go out of date fastest. Platforms change their interfaces and the way they cite sources every few months. The information below is current as of November 1, 2026, and the web version will be updated as the platforms change.

1. Same logic, different plumbing

The differences show up in four places:

• Which index the pages come from: Google’s, Bing’s or the platform’s own.
• Whether it searches on every question or only when it decides to.
• Whether sources sit inside the answer, under it or behind a button.
• Whether the site owner gets impression, citation or click data.

A study Ahrefs published in August 2025 shows how much divergence these differences add up to. Ahrefs ran 15,000 long questions through Google and Bing and also asked them to AI assistants. Only about 12% of the links cited by ChatGPT, Gemini and Copilot appear in Google’s top 10 for the same question. For Perplexity the figure is higher, at 28.6% (Ahrefs).

Ahrefs explains the gap mostly through query fan-out. Assistants don’t search the user’s sentence as written; they turn it into different searches. Each platform does that translation differently and searches a different index.

Showing up on one platform doesn’t mean you show up on another. Visibility has to be measured platform by platform; Chapter 7 covers how.

2. Google AI Overviews

AI Overviews is the short AI summary that appears above classic search results. The user runs a normal Google search and, for some searches, sees this summary above the results.

It doesn’t appear on every search. Google says AI Overviews show only when its systems determine they add something to classic search (Google Search Central). They often don’t appear for brand searches, short navigational searches or questions whose answer is already obvious.

The sources come from Google’s own search index. The rule from Chapter 2 applies exactly: to be cited in AI Overviews, a page has to be indexed in Google and eligible to be shown with a snippet, and there’s no additional technical requirement. Google also notes that AI Overviews and AI Mode can use query fan-out to run multiple searches across subtopics and data sources. Supporting links sit inside and next to the summary, and Google says these features can show a wider and more varied set of links than classic search.

Good data on which sites stand out comes from Profound, which analyzed 680 million citations collected between August 2024 and June 2025. The most-cited domains in AI Overviews were Reddit (2.2% of all citations), YouTube (1.9%), Quora (1.5%) and LinkedIn (1.3%) (Profound). The percentages are small because most citations are spread across a huge number of different sites. The names at the top, though, show how open AI Overviews is to video and forum content.

On the site owner’s side, impressions and clicks from AI Overviews are counted under the “Web” search type in Search Console. In June 2026 Google also released a separate performance report for its AI features. Open to all sites since August 31, 2026, that report currently shows only impressions, broken down by page, country, device and date; it has no click or query data (Search Engine Journal). Chapter 8 goes into the report in detail.

In Turkey, as I described in Chapter 1, AI Overviews has been rolling out gradually since February 2026. Since the launch, none of our clients has seen a serious drop in clicks, only small dips. That picture may look different depending on the industry and the type of content.

3. Google AI Mode: same index, different sources

AI Mode is a separate tab where search on Google happens entirely as a conversation. The user types a long question, gets an answer and keeps going in the same thread. Google says AI Mode is especially useful for questions that need exploration, reasoning and complex comparisons (Google Search Central).

AI Mode also draws on Google’s index. As Robby Stein described in Chapter 3, besides web results it also checks real-time data sources such as the Shopping Graph. The requirements for being cited are the same as for AI Overviews.

Same company, same index, so you’d expect the two features to cite the same pages. Data Ahrefs collected in the US in September 2025 shows they don’t (Search Engine Journal). The two features cite the same URLs for the same question only 13.7% of the time; looking only at the top three sources, the figure is 16%. The meaning of the answers, however, is very similar, with an average semantic similarity of 86%. The two features often say the same thing, but based on different sources.

The Ahrefs data shows other differences too. AI Mode answers are on average four times longer. AI Mode mentions 3.3 people or brands per answer on average, AI Overviews 1.3. Wikipedia makes up 28.9% of AI Mode citations and 18.1% of AI Overviews citations.

In practice, a question where you’re cited in AI Overviews may show other sites in AI Mode. And because AI Mode mentions more brands, it may be where the real competition happens on comparison and recommendation questions.

In AI Mode, links appear inline, next to the answer and at the end, and Google changes this area often. With changes announced in October 2026, inline links started showing higher up in the answer and more often. Deeper-dive links at the end of the answer got images and became more prominent. Google is also testing a section that shows sites considered official sources at the very top of the answer, for completing a task or getting definitive information (9to5Google).

That last test matters directly to brands. A brand’s own site is the natural official source for its products, prices and service information. If that information isn’t clear and current on your site, getting into that section becomes harder.

4. ChatGPT decides when to search

ChatGPT doesn’t search on every question, and that’s its biggest difference from Google. OpenAI’s help page says ChatGPT can search the web on its own if your question would benefit from current information, and that users can also start a search manually from the tools menu (OpenAI).

According to a Semrush study published in April 2026, based on US clickstream data, search kicked in on 34.5% of ChatGPT prompts as of February 2026 (Semrush). On roughly two thirds of prompts, ChatGPT answers from memory. As I said in Chapter 3, questions close to a sale that need current information are more likely to trigger a search.

When it does search, it uses OpenAI’s own crawler, OAI-SearchBot, and search partners. OpenAI says ChatGPT search sometimes works with other search providers and usually rewrites the query before sending it to them, but it doesn’t name the providers. In Chapter 2 we noted that ChatGPT search relied heavily on Bing data when it first launched. Today the picture is murkier. In the Ahrefs data, ChatGPT’s sources show no clear overlap with either Google’s or Bing’s top 10; only 8% of the pages it cites inline appear in Google’s top 10 for the same question (Ahrefs).

In Profound’s data, the site ChatGPT cites most, by a wide margin, is Wikipedia, at 7.8% of all citations. Reddit (1.8%), Forbes and G2 (1.1%) follow (Profound). ChatGPT leans on encyclopedic sources for definitions and general information, and on review and comparison sites for product comparisons. Wikipedia and other third-party sources are the subject of Chapter 14.

Answers that use search include inline citations, and clicking one opens the source. The Sources button under the answer lists the cited links together with other related links. In the mobile apps, some results can also include a map.

OpenAI doesn’t give site owners a search report, but it does provide two things (OpenAI). ChatGPT automatically adds the utm_source=chatgpt.com parameter to the links it gives, so you can separate ChatGPT visits in Analytics. And if you don’t want a page to appear in ChatGPT search, you can use noindex. For OAI-SearchBot to see the noindex, though, it has to be able to crawl the page; a page blocked in robots.txt can still appear as a link and title only if it’s found some other way. We covered the result of blocking OAI-SearchBot entirely in Chapter 2: you won’t be cited in ChatGPT search.

ChatGPT also has memory. It can remember a user’s earlier conversations, and that can affect the answer. Two different people asking the same question can get much more different answers than they would on Google. Shopping inside ChatGPT is covered in Chapter 15.

5. Gemini

Gemini is Google’s chat assistant. While AI Overviews and AI Mode run inside Google Search, Gemini is a separate app, and as we saw in Chapter 1, it has passed 900 million monthly users.

When Gemini needs web information, it grounds its answer in Google Search. Google’s developer documentation calls this “Grounding with Google Search.” The model analyzes the question, decides whether searching Google would improve the answer and, if so, generates one or more searches (Google AI for Developers). Like ChatGPT, Gemini decides for itself whether to search.

If the user has allowed it, Gemini can also bring files and emails from Gmail, Drive and other Google Workspace apps into the answer. Gemini’s help page says related content can come from public websites, files the user uploaded or connected Workspace documents (Gemini Help).

If there are sources, a Sources button appears under the answer or inline, and the links open in a side panel. If there’s no button, Gemini didn’t provide links for that answer.

Because Gemini uses Google Search, it’s tempting to say “if we rank well on Google, we’ll be cited in Gemini.” The Ahrefs data doesn’t support that: only about 8% of the links Gemini cites appear in Google’s top 10 for the same question (Ahrefs). Gemini uses the same index but turns the user’s question into its own, different searches.

6. Copilot and Bing

Microsoft Copilot grounds its answers in Bing’s index. That’s why I wrote in Chapter 2 that adding your site to Bing Webmaster Tools shouldn’t be treated as a nice-to-have anymore.

The Ahrefs data reflects the link too: Copilot has the highest overlap between its cited links and Bing’s top 10 (Ahrefs). Since Copilot is Bing’s own product, that’s no surprise.

In November 2025 Microsoft announced that Copilot answers would show more prominent, clickable citations. With “Show all,” users can see every source for an answer in a side panel (Microsoft). Microsoft says it made these changes with publishers and content owners in mind.

Copilot is also the platform that gives site owners the most data. The Bing Webmaster Tools AI performance report I mentioned in Chapter 2 shows how many times your pages were cited in Copilot and Bing’s AI answers, and which pages were used. It’s one of the rare cases of a search engine sharing its own citation data.

Because Bing’s search share in Turkey is small, Copilot is often ignored. On the B2B side I think that can be a mistake. Copilot is built into Microsoft’s business apps and may be one of the places corporate users ask questions during the workday. This assumption is easy to test for your own site: look at traffic from copilot.microsoft.com in Analytics.

7. Perplexity

Perplexity positioned itself as an “answer engine” from the start and put citing sources at the center of the product.

It has its own index. When describing the search infrastructure it opened to developers in September 2025, Perplexity wrote that the index covers hundreds of billions of web pages and handles tens of thousands of update requests per second. In the same post, it said the index splits documents into small pieces, scores those pieces separately against the question, and that this is the same infrastructure that powers Perplexity’s own answer engine (Perplexity). It’s a clear example of the passage-level reading from Chapter 3.

Perplexity has two bots (Perplexity). PerplexityBot crawls sites to show and link them in Perplexity’s search results; Perplexity says this bot isn’t used to train AI models and recommends allowing it in robots.txt. Perplexity-User visits a page in real time to support an answer when a user asks a question. Because the user initiates the request, it generally ignores robots.txt rules.

Every answer has numbered footnotes linking to the original sources (Perplexity), so sources are more visible than on other platforms.

In the Ahrefs data, Perplexity is the platform that overlaps most with Google: 28.6% of its sources are in Google’s top 10 for the same question. In Profound’s data, the site Perplexity cites most is Reddit, at 6.6% of all citations (Profound). In the Tow Center study from Chapter 3, Perplexity was also the tool that did best at finding the right source.

Its user base is small next to Google and ChatGPT. But because people use it for research and it shows sources prominently, it’s relatively likely to send clicks.

8. Claude

Anthropic added web search to Claude on March 20, 2025. Today web search is available on all Claude plans and in all countries. Anthropic says that when Claude uses web information, it cites sources directly so you can check them (Claude).

Anthropic doesn’t describe its search infrastructure in public product documentation. TechCrunch reported that in the week web search launched, Anthropic added Brave Search to its list of subprocessors, and that in some searches Claude and Brave returned the same sources (TechCrunch). That information is from March 2025 and is indirect evidence; we don’t know the current setup for certain.

Its bots, on the other hand, are clear. As the table in Chapter 2 showed, Claude-SearchBot handles search, Claude-User handles user requests and ClaudeBot handles training. Anthropic notes that blocking Claude-SearchBot may reduce visibility in search results.

You don’t have to guess how much your audience uses Claude; look at visits from claude.ai in Analytics. If there are few, putting this platform at the end of your measurement list is enough.

9. The same platform doesn’t give everyone the same answer

Besides the differences between platforms, there are differences within a platform. The same ChatGPT can give two users different answers to the same question.

The first reason is memory and personal context. ChatGPT can remember earlier conversations, Gemini can bring in Gmail and Drive content if allowed, and Google has taken a similar path in AI Mode with Personal Intelligence, which I mentioned in Chapter 1. Model and plan matter too. Free and paid plans can run different models, and the model a user picks, or modes like deep research, change how many searches are run and which sources are read.

Language and location have a big effect as well. A question asked in Turkish and the same question in English go to different searches and so to different sources, and the user’s country affects local results. Conversation history changes the answer too. The same question gets one answer as the first message in a chat and another at the end of a five-message conversation.

Keep this in mind when you measure. A question you ask from your own computer, on your own account, in a ChatGPT you’ve been talking to about your industry for months, may not represent the answer your customer will see. An account that asks about your brand often may also show your brand more often.

Run your checks in as clean an environment as you can: an account with memory turned off or a new one, a private window, and your target audience’s language and country. If you sell in Turkey, ask in Turkish; if you export, also ask in the language of the target market. We’ll build this into the measurement method in Chapter 7.

10. Where the link sits, and the click

In Chapter 1 we talked about the difference between being cited and being clicked. In Pew Research’s data, the click rate on links inside AI Overviews was only 1%. Part of the reason is where and how the link is shown.

Platforms design this differently. Perplexity shows sources as numbered footnotes inside the text. Copilot made sources more prominent in November 2025. Google moved inline links higher in AI Mode and AI Overviews in October 2026. In ChatGPT, sources sit as small inline citations and in a list behind a button.

The design isn’t up to you, but the sentence next to the link and the page the link goes to are. If AI takes a fact from your page and puts a link next to it, that sentence is the user’s reason to click. If it’s clear your page has something the answer can’t include, like “capacity table and technical drawings,” the chance of a click goes up; whether a page still has a job after the answer is the subject of Chapter 11. And when users click, they should see the fact they read in the answer right away. If AI took “3,000 bottles a day capacity” from your page, that fact should sit somewhere visible on the page.

11. The platforms side by side

A summary of everything so far:

The last column deserves the most thought. Google and Microsoft show you something from their own data. For ChatGPT, Gemini, Perplexity and Claude, all you have is the visits that reach your site. On these platforms you can’t see cases where you were cited but not clicked, or recommended but never linked. You have to fill that gap with your own measurement; Chapter 7 deals with that, and Chapter 8 with the Analytics side.

12. Keeping the door open on every platform

After this many differences, it’s natural to ask whether you need a separate strategy for each platform. I don’t think so. The differences are real, but the foundation for being cited anywhere is shared:

• Being indexed in Google. AI Overviews, AI Mode and Gemini rely on Google directly, and ChatGPT’s sources aren’t independent of the foundation from Chapter 2 either.
• Being indexed in Bing. Required for Copilot, useful for some other platforms.
• Keeping the door open to search bots. Make sure OAI-SearchBot, PerplexityBot and Claude-SearchBot aren’t blocked in robots.txt, at the CDN or by a security plugin.
• Putting key information in the initial HTML. As we saw in Chapter 2, some bots outside Google don’t run JavaScript.
• Being described correctly off your site too. ChatGPT leans on Wikipedia, Perplexity and AI Overviews on Reddit, AI Overviews on YouTube. If your information on those sources isn’t consistent, each platform may describe you differently.

The differences show up mostly in priorities and measurement. Take a filling machine manufacturer whose customers are purchasing managers, production managers and business owners. For this manufacturer, AI Overviews and AI Mode are probably where it reaches the widest audience. ChatGPT is where decision makers ask their long, constraint-heavy questions. Copilot is where a purchasing specialist working with Microsoft tools on a company computer might come across it. Rather than guessing which matters most, measure it.

In terms of order, starting with Google makes sense, because search in Turkey still happens mostly on Google, and AI Overviews and AI Mode are now live here. ChatGPT comes next: it’s the chat assistant with the most users and the platform whose behavior differs most from Google. Rank the rest by your audience. Look at which platforms send traffic in Analytics and which tools your sales team hears customers mention; a “How did you hear about us?” question on the quote form helps here too.

In Chapter 1 I wrote that, for our clients, AI traffic usually stays under 10% of total traffic but converts at around 6% to 7%. How that traffic splits across platforms depends on the site and the industry. The most reliable way to set priorities is to look at that split in your own data.

13. What this means for your measurement plan

AI search is made up of several channels that look alike but aren’t the same. They all work on the same logic, but they read from different indexes, search at different times and put links in different places. Even Google’s own two features cite the same URLs for the same question only 13.7% of the time.

The foundation is shared across platforms. Being indexed, staying open to bots, writing information clearly and being consistent off your site help everywhere. Measurement, though, has to happen platform by platform. Google and Microsoft show you something; elsewhere you have no data except your own measurement.

This field has also picked up a pile of new names in the last two years: GEO, AEO, LLMO. Chapter 5 sorts out what behind those names is actually new and what is just a new label.

Checklist

Sources

• Google Search Central: AI features and your website (last updated December 10, 2025)
• Search Engine Journal: Google Search Console AI reports rolled out worldwide (reporting Google’s August 31, 2026 global rollout note)
• Search Engine Journal: Google AI Mode and AI Overviews cite different URLs, per Ahrefs report (December 15, 2025)
• 9to5Google: Google AI Mode and AI Overviews link changes (October 15, 2026)
• Ahrefs: Only 12% of AI cited URLs rank in Google’s top 10 for the original prompt (August 11, 2025)
• Profound: AI platform citation patterns (June 5, 2025, updated August 2025)
• OpenAI: Searching the web with ChatGPT (help page, checked November 1, 2026)
• OpenAI: Publishers and developers FAQ (help page, checked November 1, 2026)
• Semrush: ChatGPT search insights (April 7, 2026)
• Google AI for Developers: Grounding with Google Search (checked November 1, 2026)
• Gemini Help: View related sources and double-check responses (checked November 1, 2026)
• Microsoft: Bringing the best of AI search to Copilot (November 7, 2025)
• Perplexity: Introducing the Perplexity Search API (September 2025)
• Perplexity: Perplexity crawlers (checked November 1, 2026)
• Perplexity: How does Perplexity work? (checked November 1, 2026)
• Claude: Web search (March 20, 2025, with update note)
• TechCrunch: Anthropic appears to be using Brave to power web searches for its Claude chatbot (March 21, 2025)

Previous: Chapter 3, How Does AI Build an Answer? · Table of contents · Next: Chapter 5, GEO, AEO, LLMO: What Has Really Changed?