How to get recommended in ChatGPT, Gemini and Perplexity answers

Each major assistant works a little differently, and the differences shape the tactics.

ChatGPT answers from a mix of model knowledge and live web search. When search is triggered, it retrieves a small set of pages, reads passages, and composes an answer with citations. Shopping queries can also pull structured product data from merchant feeds.

Gemini sits on top of Google's index and shares much of its retrieval with Google Search, including the signals behind AI Overviews. Strong Google presence carries over more directly here than anywhere else.

Perplexity searches the live web for nearly every query and cites heavily. It is the most responsive to on-page changes and the easiest place to check whether your fixes worked.

What they share: they retrieve fragments rather than whole pages, they reason about entities rather than keywords, and they weight sources that other credible sources also mention. A business described the same way across its own site, Google Business Profile, LinkedIn, industry directories and review platforms is far easier to recommend than one with three different descriptions and an old address.

Before changing anything, build a baseline. This is panel research, not rank tracking.

1. Write 30 to 60 questions a real US buyer would ask. Include direct ones ("best commercial HVAC contractor in Phoenix"), comparison ones ("X vs Y for small law firms") and problem-led ones ("how do I reduce cart abandonment on Shopify").
2. Ask each question in ChatGPT, Gemini and Perplexity from a US location. Use fresh sessions so prior conversation does not bias the answer.
3. Log three outcomes per question: named with a citation, named without a citation, or absent. Record which competitors were named when you were not.
4. Repeat monthly. Answers are non-deterministic, so look at the share of questions where you appear over a quarter rather than any single run.

Typical starting points for US small and mid-sized businesses: appearing in well under a quarter of relevant answers, usually only on branded questions. Treat that as a floor, not a failure.

Most invisible businesses are not being rejected. They are not being seen.

• Check the raw HTML. Fetch your key pages without JavaScript and confirm the text is present. Content rendered only client-side is invisible to many retrievers.
• Audit robots.txt by crawler. OpenAI, Google, Anthropic and Perplexity each use named user agents, and some separate training crawlers from live search crawlers. A blanket block added years ago may be excluding you from answers while leaving Google Search alone. Decide per crawler and write the decision down.
• Test for bot challenges. CDNs and firewalls often serve a challenge page to non-browser user agents. Your customers see the site; the assistant sees a wall.
• Fix indexability basics. Consistent canonicals, no accidental noindex, an honest sitemap, pages that load in under three seconds on a mobile connection.

This is build work more often than marketing work, which is why our website design and development team is involved in nearly every engagement of this type.

Once you are retrievable, structure decides whether you are used.

Answer first. Under each heading, state the answer in one or two self-contained sentences, then qualify and explain. The build-up-then-conclusion structure most writers use is backwards for retrieval.

Match headings to questions. "How much does a commercial roof replacement cost in Texas" beats "Our Pricing Philosophy."

Make sections stand alone. No "as mentioned above," no pronouns that depend on an earlier paragraph. If a passage is lifted out, it should still make sense.

Be specific and checkable. Service areas by city and state, price ranges in dollars, turnaround times, what you do not do. Vague copy is unquotable because there is nothing to quote.

Add real FAQs. Forty to seventy words each, each one complete. These are the easiest passages on the web to retrieve.

Create comparison and category pages honestly. Assistants love "X vs Y" and "best options for Z" pages because they map directly to how people ask. Write them with real criteria, name competitors fairly and make clear where you fit. Misleading comparisons are an FTC problem as well as a trust problem.

Confident attribution is what produces a recommendation. Give the assistant a clean, consistent picture.

• Organization schema on the site with the same legal name, description, address, phone and social URLs everywhere.
• A thorough About page, named leadership, and author pages for anyone writing substantive content, with credentials that explain why they would know.
• A Google Business Profile that matches the site exactly, with categories, services and hours maintained.
• Identical name, address and phone across Apple Business Connect, Bing Places, Yelp, BBB, industry directories and LinkedIn.
• Visible published and updated dates on content pages.

Brand consistency here is not cosmetic. If four sources describe you four different ways, an assistant has no confident reason to choose you. This is where brand strategy and identity work shows up in a growth result: one clear statement of who you are, used everywhere.

Assistants weight what others say about you more than what you say about yourself. The work is ordinary and slow.

• Get listed and accurately described wherever your category is discussed: industry publications, local news, trade associations, "best of" lists with real editorial standards.
• Collect reviews on Google and the platforms your buyers actually use. Respond to them. Volume and recency both matter, and so does the FTC's rule on fake reviews; never incentivize positive reviews or suppress negative ones.
• Participate in communities where your buyers ask questions, under a real name, with genuinely useful answers. Reddit and niche forums are heavily retrieved by several assistants.
• Publish original data or a point of view that other sites want to cite. One useful benchmark report earns more entity authority than fifty generic blog posts.
• Avoid volume plays. Mass directory submissions and paid placements on low-quality sites add noise to your entity rather than authority.

If you sell products, ChatGPT and Gemini can surface them from structured feeds rather than from your pages. Keep your Google Merchant Center feed complete and accurate, with current prices, availability and shipping to US addresses. Watch for merchant programs from OpenAI and others and join when they open to your category. Ensure product pages carry full Product schema with price, availability, reviews and GTINs.

Set targets you can actually observe.

• Share of panel questions where you are named, by assistant, tracked quarterly.
• Referral sessions from chatgpt.com, perplexity.ai and gemini.google.com in GA4, judged on conversion rate rather than volume. These visitors typically convert at a higher rate than general organic because they arrive pre-qualified.
• Branded search volume and direct traffic trend, since many assistant users search your name afterward instead of clicking a citation.
• Lead source field in your CRM with "AI assistant" as an option, because a surprising number of buyers will tell you if you ask.

Waiting for a tool to solve it. AI visibility trackers are useful for monitoring, but none of them change your pages or your reputation.

Treating it as a separate channel. The same crawlability, structure and entity work serves Google, AI Overviews and every assistant at once. Running parallel programs duplicates effort.

Publishing more rather than clearer. Thin additional pages dilute your entity. One page that states a specific thing precisely beats five that circle it.

Blocking crawlers by default. Check robots.txt and CDN rules today, and again after any infrastructure change.

Skipping the baseline. Without a first panel run, you will never know whether the work did anything.

We run this as one connected program rather than an SEO add-on. Discovery establishes the question panel, audits retrievability crawler by crawler, and maps every place your business is described on the web. Planning sets the target: which questions you should be winning, which pages need restructuring, which entity gaps to close and which third-party sources to pursue. The build phase is shared between teams: web fixes rendering and crawler access, brand tightens the single description used everywhere, growth restructures pages and runs the outreach, and automation wires referral sources and lead-source fields into your CRM so you can see results. Then we improve on a monthly panel and a quarterly review, moving effort toward the assistants and questions that are producing qualified leads. Our digital strategy and advisory practice leads this work.

A representative pattern is our Brand & Digital Launch engagement, which took a business from positioning through identity, website and campaign assets as one connected project. The relevance to assistant visibility is the sequence: a single clear positioning statement was defined first, then expressed identically in the site copy, the structured data, the profiles and the launch content. Because the entity was consistent from day one and the site was built to be read without JavaScript, the business could be retrieved and attributed cleanly across search and assistant surfaces rather than needing a cleanup project later.