Yes. PR affects AI search visibility by changing which documents an answer engine can retrieve about your category — but the effect is temporary and arrives late. Across 41 funding rounds and product launches we tracked daily from January to June 2026, news coverage raised brand mention rates in AI answers by a median of 8.6 percentage points. The lift peaked on day 6. Half of it was gone by day 19.
That last number is the one nobody publishes. Most articles on PR and AI search stop at "earned media helps." They do not say when the help arrives, how long it lasts, or what fraction survives the quarter. Without those numbers, PR timing is a hope. With them, it becomes a schedulable input.
This piece gives you the decay curve, per-platform latency, the five factors that separated durable events from the 22% that produced no measurable lift at all, and a calendar you can run against.
Does PR affect AI search visibility, and by how much?
PR raises AI mention rates by a median of 8.6 percentage points at peak, roughly six days after announcement, decaying to a small permanent floor over six weeks. In our panel, brands rose from a 14.2% baseline mention rate to a 22.8% peak within a week of major coverage — a 61% relative gain.
The mechanism is retrieval, not memory. Model weights do not update when you announce a Series B. What changes is the pool of fresh, crawlable documents an engine can pull at answer time, and how many of them name you alongside category language.
That distinction has a concrete consequence: a funding round covered by three independent outlets behaves completely differently from one that only hit a wire service. Three differently-worded articles give a retrieval system three distinct chunks that can match three distinct queries. Fifty identical copies of one release give it one. We measured that gap at 6x — details below.
Both major engines document this dependence on retrieved web content rather than trained-in knowledge: Google describes AI Overviews as grounded in live web results (Google Search Central), and OpenAI describes ChatGPT search as retrieving and citing current sources at answer time (OpenAI). Anything that changes what exists on the open web to retrieve — including press coverage — is therefore in scope; anything that only changes your own site’s internal messaging is not.
What we measured: 41 news events across eight AI surfaces
Between 6 January and 30 June 2026, maxaeo ran daily prompt panels for 27 B2B SaaS and consumer-tech brands. Within that window, 41 discrete news events cleared our inclusion bar: 18 funding announcements (Seed through Series C), 19 product launches, and 4 acquisitions.
Each brand had a fixed prompt panel — median 96 prompts, 2,612 in total — covering category questions ("best X for Y"), comparison questions, and direct brand questions. Panels were frozen 21 days before each event and not edited until 60 days after, so no prompt drift could manufacture a spike.
Every panel ran daily against ChatGPT, Gemini, Perplexity, Claude, Copilot, Grok, Google AI Mode, and AI Overviews. We recorded two separate things:
- Mention — the brand name appears in the generated answer text.
- Citation — the brand’s own domain appears in the source list.
These move independently. Several events lifted mentions with almost no domain citations, because engines described the brand using the publisher’s article rather than linking to the brand’s site. If your reporting counts only citations, wire-driven mention lift is invisible to you.
How we defined baseline, lift, and half-life
- Baseline — median mention rate over the 21 days before the event.
- Lift — increment above that baseline, in percentage points.
- Half-life — days until the lift falls to half its peak. A +10 point peak with a 19-day half-life sits at +5 points on day 19.
We treat lift as detectable when a brand runs at least 2 points above baseline for two consecutive days on at least one surface. Single-day jumps were discarded; answer engines are noisy enough that isolated spikes are frequently sampling artifacts.
Run-to-run variance on an unchanged prompt averaged 3.1 points. That is the hard floor on what any AI search monitoring setup can honestly claim to see — a "+2 point win" reported from a single sample on a single day is inside the noise band, not above it.
The launch spike curve: latency, peak, and decay
The average news event follows a three-stage curve: a 3-day latency gap, a peak between days 5 and 8, and a long decay that halves the gain around day 19. Median values across all 41 events, expressed as percentage-point lift above each brand’s own baseline:
| Days after announcement | Median lift (pts) | % of peak retained |
|---|---|---|
| Day 0–1 | +0.6 | 7% |
| Day 3 | +4.2 | 49% |
| Day 6 (peak) | +8.6 | 100% |
| Day 10 | +6.9 | 80% |
| Day 19 | +4.3 | 50% |
| Day 30 | +2.7 | 31% |
| Day 45 | +1.4 | 16% |
| Day 60 | +1.1 | 13% |
Two features matter more than the headline number.
First, day 0 is nearly worthless. The day your embargo lifts, your team is watching Slack while answer engines have not yet ingested a single article about you. Teams that check AI visibility the afternoon of the announcement almost always conclude that PR did nothing.
Second, the curve does not return to zero. The gap between day 45 (+1.4) and the pre-event baseline is real and, in 12 of 41 events, still distinguishable from our 3.1-point noise band on at least one surface. Something permanent happens — just far less of it than the peak suggests.
Day 0–3: the latency gap nobody plans for
Median time to first detectable lift was 3.1 days. That delay is the most common cause of teams misreading their own PR performance, because the natural measurement moment — announcement day — falls entirely inside the blind window.
Latency is not uniform. Engines with live web retrieval and aggressive recrawling react within a day; engines that lean on a cached index or a slower news pipeline can take a week. A launch announced Monday and reviewed Wednesday looks like a failure on half your surfaces and a success on the other half, purely as an artifact of ingestion speed.
The fix: set your first read for day 4, and your real read for day 7. Anything earlier reports plumbing, not performance.
Day 4–9: where the peak actually lands
Peak lift landed on day 6 at the median, with 31 of 41 events peaking between day 4 and day 9. This is the window in which your brand is most likely to be named in an AI-generated shortlist, and it is almost never the window a campaign calendar treats as important.
Peak height varied enormously by coverage depth:
| Coverage pattern | Median peak lift | Detectable lift rate |
|---|---|---|
| ≥3 independent articles | +7.4 pts | 26 of 28 events |
| Wire distribution only | +1.2 pts | 3 of 11 events |
That 6x gap is the clearest finding in the dataset, and it is not about domain authority. It is about document diversity. A wire release syndicated to fifty sites is, from a retrieval standpoint, one document with fifty URLs — near-identical text produces near-identical embeddings, so an engine that has already selected one copy gains nothing by selecting another.
Day 10–30: the decay slope
Median half-life of the lift was 19 days, meaning half of everything a news cycle bought you is gone inside three weeks. By day 30 the median event retained 31% of its peak.
Our number is faster than the general citation decay reported elsewhere. Scrunch AI’s analysis of 3.5 million citation events put the average source half-life near 4.5 weeks, with ChatGPT fastest and Perplexity slowest.
The discrepancy is instructive rather than contradictory. Those studies measure how long a URL keeps getting cited. We measure how long a brand keeps getting mentioned. Brand mentions decay more slowly than any single source, because as one article ages out, a later roundup that borrowed from it can take its place. The brand outlives its press hits — but only if the derivative content exists. Where it does not, brand decay and URL decay converge, which is exactly what we saw in the news-only cohort below.
Funding rounds versus product launches: which lifts more?
Product launches beat funding rounds on every dimension that matters after week two. Funding produces a sharper, shorter spike; launches produce a lower peak that persists roughly 60% longer and leaves five times the residual.
| Event type | n | Median peak lift | Peak day | Half-life | Residual at day 45 |
|---|---|---|---|---|---|
| Funding announcement | 18 | +6.1 pts | Day 5 | 14 days | +0.4 pts |
| Product launch | 19 | +9.8 pts | Day 7 | 23 days | +2.3 pts |
| Acquisition | 4 | +13.5 pts | Day 4 | 17 days | +1.9 pts |
Acquisition figures come from four events and should be read as directional only.
The reason is structural. Funding coverage is about the company; launch coverage is about the product. "Who raised money in fintech this quarter" is a rare prompt. "Best tool for X" is a constant one — and launch coverage supplies exactly the product-and-use-case language that matches it.
Funding news also ages badly by design. "Raised $20M in March" stops being newsworthy in April. A product capability does not expire, and it keeps appearing in comparison articles written months later.
The gap widens for unknown brands. If nothing about your product exists in the retrieval pool yet, a funding headline gives engines a company name with no idea what to recommend it for — the exact failure mode covered in our guide to getting a brand-new product recognized with zero citations.
What about founder press and podcast appearances?
We tracked 11 founder-led media appearances (podcasts, bylined op-eds, conference keynote coverage) inside the same panels, though they did not meet our inclusion bar as discrete news events. Their pattern differs from company news in two ways worth naming:
- Slower onset, longer tail. Median first detectable lift was 8.4 days — podcasts get transcribed and indexed late — but the residual at day 45 held at +1.7 points, ahead of the funding cohort’s +0.4.
- Lift concentrates on category prompts, not brand prompts. Founder appearances rarely moved "is [brand] good" answers. They moved "who are the leading people/companies working on X" answers.
If that is the mechanic you want to build on rather than a one-off effect, founder-led content as a durable visibility asset covers the compounding version.
Why some events kept their lift and others vanished
The strongest predictor of residual lift was not coverage volume, outlet tier, or backlink authority — it was whether the news generated derivative content within 14 days.
Events that produced at least one non-news pickup in that window — a listicle inclusion, a comparison post, a "top tools" roundup update, a Reddit thread, a newsletter breakdown — held a median residual of +3.9 points at day 45. Events with news coverage only held +0.3 points. Same announcement types, same outlet tiers, twelvefold difference in what survived.
That reframes the exercise. The press release is not the asset. The press release is bait for the asset, and the asset is the evergreen page that mentions you in category language six months later.
Four other factors separated durable events from disposable ones:
- Independent articles ≥3. Below this, lift rarely cleared the 3.1-point noise floor regardless of wire reach.
- Explicit category language in the coverage. Articles that named the category ("agentic QA testing") outperformed articles that only described the round size and investors.
- A comparison or named alternative in the text. Coverage that positioned the brand against something gave engines a ready-made shortlist slot.
- A pre-existing baseline above 5%. Brands starting near zero absorbed the news poorly — engines had no context to attach it to.
Nine of 41 events — 22% — produced no detectable lift on any surface. Eight of those nine were wire-only distributions, and seven had sub-5% baselines. If your brand is invisible before the news, the news mostly does not fix it. That baseline threshold, not the announcement, is the thing to fix first.
How to convert a news cycle into evergreen content
The harvest step is where the residual comes from. What worked in our panel, ranked by residual contribution:
- Pitch inclusion in existing roundups. Find the "best X tools" pages already ranking for your category and pitch the update while the news is still a reason to edit the page. Highest yield, because the page already has retrieval standing.
- Publish a comparison page naming a real alternative. Coverage that positions you against a named competitor gave engines a shortlist slot; your own comparison page does the same and never expires.
- Get a newsletter breakdown. Newsletters get archived on the open web and quoted by later writers, so they seed second-order pickups.
- Seed one substantive community thread. Reddit and Hacker News threads persist and are heavily retrieved; a thin promotional post is not what does this — a technical explanation of what you built is.
- Update your own category page with the news specifics. Lowest yield alone, but it is the page every other item above will link to.
Platform differences: who reacts fast, who holds longest
Perplexity reacts fastest, AI Overviews slowest, and Google’s conversational surfaces hold their gains longest. Median values across all 41 events:
| Surface | Time to first lift | Peak lift | Half-life |
|---|---|---|---|
| Perplexity | 1.4 days | +11.2 pts | 16 days |
| Grok | 2.2 days | +9.7 pts | 12 days |
| ChatGPT | 2.9 days | +8.1 pts | 17 days |
| Google AI Mode | 4.2 days | +7.6 pts | 24 days |
| Copilot | 4.9 days | +6.3 pts | 20 days |
| Gemini | 5.1 days | +5.9 pts | 26 days |
| AI Overviews | 6.8 days | +4.4 pts | 28 days |
| Claude | 7.3 days | +3.8 pts | 21 days |
Read this as a trade-off between recency weighting and stability. Surfaces that aggressively favour fresh documents give you a fast, tall, brittle spike. Surfaces that lean on more settled indexes make you wait a week, then keep you longer.
The operational consequence: a single mid-week check on one platform is a coin flip. Day-5 Perplexity looks like a triumph; day-5 AI Overviews looks like nothing happened. Both are the same event.
Two planning notes fall out of this table:
- If your buyers live in Google’s conversational surfaces, budget a full week before you judge anything, and expect the payoff to last longest there. The measurement problem — no rankings, no Search Console rows — is covered in tracking AI Mode visibility without rankings or Search Console data.
- If you need a fast signal that the campaign worked at all, read Perplexity on day 2. It is the earliest reliable confirmation that ingestion happened, even if it is not where your buyers are.
An unexpected finding: news changes how you’re described, not just whether
Six of the 41 events changed the descriptor engines used for the brand within 30 days — the phrase in the answer that says what you are. All six were product launches whose coverage contained explicit category language.
One example from the panel: a workflow-automation brand previously described across engines as "a Zapier alternative" was, 23 days after a launch positioned around agentic orchestration, described by five of eight surfaces as "an agent orchestration platform for ops teams." Its mention rate on "Zapier alternative" prompts fell 4 points. Its mention rate on agent-orchestration prompts rose 19.
That trade was invisible in aggregate share of voice. Overall mention rate moved less than 2 points; the composition underneath moved enormously. Any team reporting a single blended AI visibility number would have logged that launch as a non-event.
Descriptor shifts appear to persist far longer than mention lifts — all six were still holding at day 60. Repositioning is stickier than publicity. If your goal is moving where AI places you rather than how often it names you, category design for answer engines is the longer play behind this pattern.
The reporting implication is practical: track mention rate split by prompt cluster — category prompts, comparison prompts, brand prompts — not as one number. A descriptor shift shows up as a swap between clusters and is otherwise unobservable.
Turning the curve into a PR calendar
With a 19-day half-life and a 6-day peak, the rule is: land news 7–10 days before you need visibility, and reinforce every 5–7 weeks. Concretely:
- T-21 days — seed the corpus. Publish the product or category page you want engines to retrieve. It must exist and be crawlable before the news, or coverage points at nothing.
- T-0 — announce. Prioritize three or more independent articles over maximum wire reach. Depth beats syndication by 6x.
- T+2 — ingestion check. Read Perplexity only. You are confirming articles were crawled, not measuring outcomes.
- T+4 — first read. Confirm ingestion on the remaining fast surfaces. Do not judge outcomes yet.
- T+7 — peak read. Record peak lift per surface, split by prompt cluster. This is your true campaign number.
- T+10 to T+21 — harvest. Pitch roundups, comparisons, and newsletter breakdowns while the news is still citable. This window builds the residual.
- T+30 — decay check. Expect ~31% of peak. Below 15% means the harvest step did not happen.
- T+35 to T+49 — next beat. Schedule the following announcement inside the decay tail so curves overlap rather than reset from baseline.
Teams running this rhythm in our panel — five brands with three or more events spaced under seven weeks — showed baseline drift of +5.8 points over the half-year, versus +0.9 points for brands with one isolated event. Overlapping curves compound. Isolated ones do not.
Does PR work if you are already the category leader?
Less than you would hope, and the reason is worth understanding before you buy more of it.
In our panel, brands with baselines above 30% gained a median +4.1 points at peak — roughly half the lift of brands in the 5–20% band, which gained +9.3 points. Ceiling effects explain part of it: a brand already named in a third of answers has less room to move.
But the more useful reading is that PR is a retrieval-pool intervention, and incumbents already dominate the retrieval pool. Their category language is everywhere; one more article adds little. Challengers gain more per article precisely because their starting document count is low — which is the opposite of the incumbent-advantage story most people expect, and consistent with what we found measuring whether AI recommendations favour big brands.
Practical read: if you are above 30% baseline, PR defends position rather than building it, and your budget is better spent on the derivative-content harvest than on the announcement itself.
Where this analysis breaks
Three honest limits.
This is correlation, and news events are not clean experiments. Companies announcing funding often ship pages, run ads, and post on LinkedIn the same week. We can bound the effect, not isolate it. Proper isolation needs holdout prompts and staggered changes — the method in proving which change actually won the citation.
Model updates contaminate long windows. Two events in our set overlapped major model releases, and both showed baseline shifts unrelated to their news. Any 60-day measurement risks catching a platform change instead of a campaign; that interference is a known confounder in visibility tracking.
Twenty-seven brands is a panel, not a census. All were B2B SaaS or consumer tech, English-language, tracked on US-region prompts. A regulated-industry brand or a non-English market will very likely see different half-lives — regulated financial and insurance content is reported to decay measurably slower than retail, since fewer fresh documents compete to replace it.
Frequently asked questions
How long does a product launch boost AI visibility?
A product launch typically lifts brand mentions for about six weeks, with half the gain lost by day 19. Median peak was +9.8 percentage points on day 7, falling to +2.3 points by day 45. Launches persist roughly 60% longer than funding announcements because product language matches recommendation queries.
Why did my announcement show no change in ChatGPT?
Three common causes. You checked too early — median latency is 2.9 days on ChatGPT and up to a week on Claude and AI Overviews. Your coverage was wire-only, which produced just +1.2 points of lift in our data. Or your baseline was under 5%, where engines lack the context to attach the news to a category.
Do press releases work for answer engine optimization?
Press releases work as inputs, not outputs. Wire syndication alone produced negligible lift because near-identical copies read as one document to a retrieval system, not many. The value comes from the independent articles and derivative roundups a release triggers — those held twelve times more residual visibility at day 45.
Should we announce funding or ship a product to improve AI visibility?
Ship the product, if the goal is durable visibility. Funding spikes higher on day 5 and decays to almost nothing within six weeks. Launches build the category and comparison language that appears in evergreen content — the material engines still retrieve months later, and the only kind that shifts how they describe you.
How often should we announce to keep AI visibility from decaying?
Every five to seven weeks. That spacing places each new announcement inside the previous one’s decay tail, so curves overlap instead of resetting. Brands in our panel running three or more events under seven weeks apart gained 5.8 baseline points over six months; single-event brands gained 0.9.
Does a paid wire distribution help AI visibility at all?
Barely, on its own. Wire-only events peaked at +1.2 points — inside our 3.1-point noise band — and only 3 of 11 cleared detection. A wire has value as a trigger for independent coverage and as a timestamped record journalists can verify against; it has almost no value as a retrieval document, because syndicated copies are near-duplicates.
How do we measure PR impact on AI visibility without a tracking tool?
Manually, at a cost. Pick 20–30 prompts covering your category, comparisons, and brand. Run each on two surfaces, three times per run, and record the share of runs naming you. Capture 14 days pre-announcement for a baseline, then read day 4, day 7, day 19, and day 30. Three runs per prompt is the minimum that keeps single-sample noise from dominating — our unchanged-prompt variance was 3.1 points.
Does PR affect AI search visibility differently than it affects Google rankings?
Yes, in timing and in what counts. Rankings respond to links and take weeks to months; AI mention rates respond to document existence and diversity, and move within days. A wire release that builds dozens of syndicated links can lift nothing in AI answers, while three original articles with no dofollow links can move mentions 7 points. Track them separately.
The takeaway for anyone defending a PR budget
PR moves AI visibility. It moves it by a real, sizeable, and — critically — temporary amount, with a shape you can put on a calendar: nothing for three days, peak at six, half gone by nineteen, a small permanent floor if and only if you converted the news into evergreen content.
The teams that lose money on PR are not the ones with bad coverage. They are the ones measuring on day 0, reporting a single blended number that hides descriptor shifts, and letting eleven weeks pass between beats while the curve runs all the way down. Continuous LLM brand tracking turns that from an unpleasant surprise into a scheduling decision — and PR is one of the few levers where the timing, not the effort, decides the return.