How to show up in ChatGPT and the other AI engines

Last updated October 14, 2026. Written by the Third Coast AI team.

To show up in ChatGPT, Google AI Overviews and AI Mode, Gemini, Perplexity and Copilot, you need to do six things in order: find out what the engines currently say when your buyers ask, capture the hidden sub-queries behind those questions, learn the decision factors the engines use to pick a winner, put proof of each factor on your own pages in a form a machine can quote, build the third-party record that lets the engine trust what you say, and re-run the questions to see whether the answer changed. This is the method Third Coast AI runs for clients, written out so you can run it yourself or know what you are buying.

There is no trick in it, and the parts that work are the parts backed by published research: Aggarwal and colleagues at Princeton (KDD 2024) measured quotations, statistics and cited sources lifting visibility in generated answers by about 41%, 33% and 28%, and Ahrefs' 75,000-brand study found third-party mentions correlate with AI citations three times as strongly as backlinks. What follows is how to turn those findings into a Tuesday afternoon's work.

Where should you start?

Start by treating every AI answer as a mini-RFP. When a buyer asks ChatGPT "who should I hire for X," the engine is running a small procurement process: it decides what the criteria are, it gathers candidates, it scores them against the criteria using whatever it can read, and it writes a recommendation. Your job is the same as answering any RFP: find out the criteria, supply evidence for each one, and make sure your references check out. Everything below is a version of that.

1. Build a prompt bank. Write 60 to 120 prompts the way your buyers actually ask, across four types: visibility ("best [category] in [place]"), decision-factor mining ("what should I look for in a [category]?"), competitor ("compare [you] and [rival]") and fact check ("does [you] offer [thing]?"). Write them for personas, not keywords.
2. Run them with retrieval on. ChatGPT with web search enabled, Gemini with grounding, plus AI Overviews, AI Mode, Perplexity and Copilot. Run each prompt more than once; a single snapshot is unreliable because answers vary by phrasing and session. Save the answer, the citations and, where you can capture them, the hidden searches the engine issued.
3. Extract the decision factors. From the decision-factor prompts, list the criteria the engines use for your category in their own words and count how often each appears. This list is your RFP.
4. Audit your pages for proof. For each factor, find the sentence on your site that states it plainly and backs it with a number, a named source or a verifiable claim. Most companies find they state a third of the factors and prove fewer.
5. Read the pages the engines cite. For each topic, open the cited pages and note where the answer sits, how many statistics appear early, whether there is a table, how the headings are phrased, and what they all say that you do not. Write your brief from that, not from a keyword tool.
6. Write and publish to a gate. Answer in the first 150 to 200 words, state and prove the factors, include sourced statistics, an attributed quote, a list and a table, phrase headings as buyers ask, and put it all in the raw HTML. Invent nothing.
7. Fix the record. Reconcile every profile, create a Wikidata item, close the schema sameAs chain, correct what the fact-check prompts showed the engines getting wrong, and earn mentions on the sources the engines cited for your category.
8. Re-run and keep going. After indexing, re-run the prompt set and log the change at the prompt level. Repeat monthly, refresh the pages that gained citations, and move on to the next cluster.

Why do the engines' hidden searches matter more than the keyword?

Because the engine does not search for what the buyer typed. Google has described AI Mode as issuing multiple related queries for one prompt, and our captures routinely show 10 to 20 sub-queries behind a buyer question: for "best swim school for a nervous four-year-old," the engine goes looking for instructor ratios, water temperature, parent reviews, class size, trial lessons, and nearby alternatives. A page optimized for the exact phrase matches one of those. A page with a self-contained passage for each matches several and gets cited. BrightEdge measured pillar-and-cluster architectures lifting AI Overview citation on pillar queries from about 12% to 41% in the first quarter of 2026, which is the same finding at site scale.

Which page elements make the biggest difference?

Two cautions on reading that table. The GEO paper's experiments used 2024-era models, and nobody has fully replicated them on the current generation, so treat the percentages as direction rather than promise. And the last two rows matter as much as the first: the engines penalize the shortcuts, and Google has issued manual actions against self-promotional listicles and against sites that published AI-generated pages at scale.

Do ChatGPT, Gemini and Perplexity want the same things?

Mostly, with differences that change the order of work. ChatGPT's retrieval does not execute JavaScript, so anything that matters must be in the initial HTML, and Wikipedia is its single largest source at about 7.8% of citations, which makes entity work and encyclopedic clarity unusually valuable there. Perplexity cites 16 or more sources per answer at low influence each and favors fresh content (half its citations are under 13 weeks old), so breadth and recency pay. Google's AI Overviews and AI Mode lean on Reddit, YouTube and topical clusters and are fed by the normal Google index, so classic SEO health matters most there. Copilot is fed by Bing, uses schema for grounding, and now offers a free AI Performance report in Bing Webmaster Tools, which every company should verify. Only about 11% of cited domains overlap between ChatGPT and Perplexity, so a plan that checks one engine and assumes the rest is guessing.

Can you do this without an agency?

Yes, if you have a writer and a few hours a week. The prompt bank and the runs can be done by hand in the engines themselves; tools like PromptWatch, Profound and OtterlyAI (which publishes pricing from about $29 to $489 a month (checked October 2026)) automate the tracking; and the proof audit is a spreadsheet with your factors down the side and your pages across the top. The parts that are hard to do alone are the citation reads, which take judgment, and the gate, which requires someone willing to send a page back. If you want the system set up and handed over, that is what our training option is for, and if you want it run for you, start with the audit.

Which mistakes undo the work?

• Blocking AI crawlers in robots.txt or through a CDN's bot protection, which silently removes you from retrieval. Allow GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Bingbot, and test with raw requests.
• Rendering the important text with JavaScript. ChatGPT will not see it.
• Writing "we're the best" pages. Google's January 2026 manual actions against self-promotional listicles were tracked publicly by Lily Ray.
• Changing the date on a page without changing the content. Header-only refreshes are detected.
• Measuring once. Run every prompt more than once and report a rate.
• Optimizing for one engine. Citation sources barely overlap between them.
• Inventing a statistic. The engines cross-check, and a fact-check prompt will eventually surface it.

Frequently asked questions