Is AI Lying About Your Brand? Defensive SEO Explained

A prospect asks ChatGPT what your company does. It confidently describes a product you discontinued in 2022, attributes a competitor's tagline to you, and rounds things off with a pricing model you've never used. The prospect nods, takes it as gospel, and moves on. You never knew the conversation happened.

That is the problem defensive SEO exists to solve. It is the work of controlling your brand narrative inside AI search, so that ChatGPT, Perplexity, Gemini and Google's AI Overviews describe you accurately instead of inventing a version of you that quietly costs you customers.

We run SEO and AI search campaigns for a living, and we'll be blunt about something most agencies won't admit: a lot of the panic around this topic is overblown, and a lot of the actual risk is ignored. This piece sorts the real threat from the noise, and tells you what we'd actually do about it.

What is defensive SEO for AI search?

Defensive SEO is the practice of shaping what AI systems say about your brand when someone asks, rather than only chasing rankings on a results page. The goal is accuracy and control: making sure the machines that increasingly sit between you and your customer get your story right.

Classic SEO was about earning the click. You optimised a page, built some links, and a human decided whether to visit. That game still matters, but a second game now runs alongside it, where the answer gets delivered before any click happens at all.

And the answer layer is enormous. According to Pew Research Center data from July 2025, around 18% of Google searches in March 2025 produced an AI summary, and when one appeared, users clicked a traditional search result just 8% of the time versus 15% when there was no summary. They clicked a link inside the summary itself in only 1% of visits.

Read that back. For a growing slice of searches, the AI's description of you is the search result. If it's wrong, you don't get a chance to correct it on your own page, because nobody arrives at your page. That is exactly why getting your brand cited and represented accurately in AI search has moved from a nice-to-have to a line item we now build into nearly every campaign.

Why AI gets brands wrong in the first place

Large language models don't look things up and verify them the way you'd hope. They predict the most statistically plausible next words based on patterns in their training data and whatever they retrieve at query time. There is no internal "let me double-check that this brand actually offers this" step.

So when the web's record of your business is thin, contradictory or out of date, the model fills the gaps with whatever pattern looks most likely. Three old blog posts mentioning a feature you killed two years ago can be enough to keep that feature alive in an AI's answer indefinitely.

It helps to understand where Google's AI features pull from, because it dismantles half the myths. Google states plainly in its official AI features documentation that "there are no additional requirements to appear in AI Overviews or AI Mode," and that to be shown as a supporting link, "a page must be indexed and eligible to be shown in Google Search with a snippet." There's no secret AI index. AI Overviews draw from the same index as normal search.

Our take: that single fact reframes the whole panic. You don't need a magic new tactic to influence Google's AI. You need the web to contain clear, consistent, current, indexable information about you, and you need that information to be the strongest signal available. Most brands we audit fail not because AI is mysterious, but because their own digital footprint is a mess of stale pages, abandoned profiles and inconsistent descriptions.

Audit what AI actually says about you

You can't fix a narrative you've never read. Before touching anything, sit down and interrogate the machines the way your customers do.

Open ChatGPT, Perplexity, Gemini, Google's AI Mode and Bing's Copilot, and run the questions a real buyer would ask. "What does [your company] do?" "Is [your company] any good?" "[Your brand] vs [competitor]." "How much does [your company] charge?" "Who founded [your company]?" Run each query two or three times, because answers wobble between sessions.

Screenshot everything. Then score each response across four dimensions. We use this exact grid with clients:

• Accuracy. Does it correctly describe your products, services, location and the basic facts of your business?
• Completeness. Are your real differentiators present, or is it describing a generic version of your category?
• Sentiment. Is the framing positive, neutral or quietly damning? An old complaint can become the headline.
• Citations. Where is it pulling from? Your own site and reputable publications, or a 2019 forum thread and a competitor comparison post?

That last column is the one people skip and shouldn't. The citations are your map. If every model keeps quoting the same outdated directory listing or a single grumpy Reddit thread, you've found the exact source you need to outweigh. This is the same entity-led thinking we cover in our guide to knowledge graphs and entity optimisation for AI search.

Run the audit monthly, not once. Models retrain, retrieval changes, and your competitors are publishing too. A snapshot tells you nothing about whether you're winning or losing ground.

What we usually find

Across the audits we run, the misfires fall into a handful of repeat offenders. Stale product or pricing info pulled from old pages you forgot to update. Confusion between your brand and a similarly named company. A founder's name missing entirely because it appears nowhere consistent. And the classic: the AI confidently inventing a feature because someone, somewhere, once wrote a wishlist post about your category.

None of that is the AI being malicious. It's the AI being lazy on your behalf, with the materials you left lying around.

Give the machines clean facts to work with

The fix starts with making the true version of your brand the easiest, clearest and most consistent thing to find. Three layers do most of the heavy lifting.

First, structured data. Organization, Product and similar schema give machines unambiguous facts: your legal name, what you sell, where you operate, who runs the place. But mark it up honestly. Google's structured data guidelines are explicit that you must not "mark up content that is not visible to readers of the page," and that markup which "is not representative of the main content of the page, or is potentially misleading" can get you penalised. Schema is a clarity tool, not a costume.

A reality check worth having, though. In its guide to optimising for AI features, Google says outright that "structured data isn't required for generative AI search, and there's no special schema.org markup you need to add." So schema helps with clarity and rich results, but anyone selling you "AI schema" as a silver bullet is selling you something Google has already contradicted. We still implement it, because clean facts never hurt, but we don't pretend it's the whole answer.

Second, consistency across the web. AI systems lean towards consensus. If a dozen credible sources describe you the same way, that description wins. If your homepage says "enterprise workflow automation," your support docs say "task manager," and an old press release says "project tool," the model blends them into mush and serves the mush back to everyone. Pin down one description and use it everywhere, from your site to your profiles to your outreach. Our walkthrough on how to get your brand into AI answers goes deeper on building that consensus.

Third, authority and freshness. Google's own AI optimisation guide warns against content that recycles "what others on the internet have already said, or could easily be produced by a generative AI model." That cuts both ways. Thin, me-too content does nothing to anchor your narrative. First-hand expertise, original data and genuinely useful pages are what get cited and quoted, which is the entire premise behind the kind of people-first SEO programmes we run.

The part nobody owns: PR, support and reviews

Here's where defensive SEO stops being a technical job and becomes an organisational one. AI doesn't only read your marketing site. It reads your reviews, your support articles, your founder's LinkedIn, your old press coverage and the forum where a customer described what you do in their own words three years ago.

When those voices disagree, the AI invents a composite that represents none of them. We've watched a client's "AI summary" get poisoned by a single detailed negative review simply because it was the most specific, recent thing the model could find. Specificity beats volume in retrieval, and that should make you nervous about what your loudest content actually is.

So the cleanup work is real work. Respond to old reviews with current facts. Update or retire ancient press releases. Publish fresh interviews, case studies and data that out-rank and out-recency the stale stuff. And prune the dead weight on your own site, which we treat as a first move, not an afterthought, in our content pruning playbook.

One more lever most people underuse: Wikipedia and other reference entities feed knowledge graphs that AI models trust heavily. Earning a legitimate, well-sourced presence there is one of the strongest ways to anchor your brand facts, which is why we wrote a whole piece on Wikipedia, brand SEO and LLM citations.

Your support team is now training data

Worth sitting with for a second. Every review reply, help article and public social response is potential fuel for how AI describes you. The tone your support team uses, the words they choose, the way they frame a fix, all of it can end up paraphrased back to a future buyer. That's not a reason to panic. It's a reason to make consistency a habit rather than a campaign.

How AI Overviews actually pick their sources

Defensive SEO inside Google specifically is less mysterious than the GEO hype suggests. Google has repeatedly tied AI feature eligibility back to ordinary search. The same documentation that says there's no separate AI index also says "the best practices for SEO remain relevant for AI features in Google Search," and warns against adding "new machine readable files, AI text files, or markup."

Translation: the page that earns a clear organic snippet is the page positioned to be cited. The structural and technical work that wins organic visibility, clean architecture, fast indexable pages, sensible internal links, is the same work that wins AI citations. If you want the granular version, our rundown of technical SEO strategies covers the foundations, and our piece on getting cited in ChatGPT and AI Overviews covers the citation-specific tactics.

The catch we keep flagging to clients: you're optimising for more than one surface. Google's AI Overviews and AI Mode don't even agree with each other, with reported citation overlap between the two sitting in the low double digits. So "getting into the AI answer" isn't one target. It's several, across several products, each pulling slightly differently.

Is this even worth the effort? The honest case

We'll resist the urge to scaremonger, because the data does the persuading on its own.

On the risk side, trust in AI answers is real but soft. Pew's October 2025 survey found that 65% of US adults at least sometimes come across AI summaries, and 53% have at least some trust in them, though only 6% trust them "a lot." So a majority will read what the AI says about you and broadly believe it, even while a chunk stay sceptical. That's enough exposure to matter and enough doubt that getting it right tips decisions your way.

On the upside, the traffic is arriving and it converts. Adobe Analytics reported in March 2025 that traffic to US retail sites from generative AI sources rose 1,200% between July 2024 and February 2025. And these visitors weren't tyre-kickers: they browsed 12% more pages per visit and bounced 23% less than other traffic. People who arrive already informed by an AI answer behave like people who've half-decided to buy. You want that answer to have been accurate and flattering, not a hallucinated mess.

So the real question isn't "is AI worth optimising for." It's "can you afford to let the most-cited description of your brand be written by a system that pattern-matches old forum posts." For most companies we work with, the answer is no.

A defensive SEO checklist you can run this quarter

Here's the sequence we'd follow if your AI narrative is a black box right now. It's deliberately boring, because the boring version works.

1. Audit five AI surfaces. Run your real buyer questions through ChatGPT, Perplexity, Gemini, Google AI Mode and Copilot. Screenshot and score for accuracy, completeness, sentiment and citations.
2. Trace the bad sources. Note every stale page, wrong directory listing or grumpy thread the models keep quoting. That list is your fix-it backlog, in priority order.
3. Lock down one brand description. Write the canonical version of what you do and roll it out everywhere, from your homepage to your profiles.
4. Clean your own house. Update or retire outdated pages, fix pricing and product facts, and add honest Organization and Product schema that matches what's on the page.
5. Out-publish the old narrative. Ship fresh, first-hand, genuinely useful content and earn citations from sources AI already trusts, reference entities included.
6. Measure monthly. Re-run the audit, track the accuracy and sentiment scores over time, and watch branded search and direct traffic as proxy signals. Better summaries make more people curious about you by name.

None of this requires a secret AI hack. It requires doing the unglamorous web hygiene that most brands have been putting off for years, then keeping it up.

If you'd rather not run the audit yourself, that's literally what our AI search visibility service exists to do, and we're happy to take a look at what the machines are currently saying about you. Tell us your brand name and a couple of buyer questions, and we'll show you the gaps before they cost you another sale.